Curated, real stories from the AI world – astonishing, absurd, sometimes unsettling. What we pick up, put in context and share. The growing archive behind our daily “Seriously?!” on the homepage.
From budget brand to $650,000 battle bot: Unitree’s mecha moment
Chinese maker Unitree, famous for cheap robot dogs and humanoids, unveiled the GD01 – a giant, transforming “mecha” robot that smashes through walls, priced at around $650,000. In the very week of its own IPO, the budget brand is putting a pure prestige piece in the window. Practical use: questionable. Viral marketing value: enormous.
“A halo product isn’t a business model – but it’s one hell of a trailer.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
86 percent: almost every US venture dollar now chases AI alone
In the first half of 2026, US startups raised about $412.7 billion – and roughly $355.9 billion of it (86%) went to AI companies. It’s the most extreme concentration ever recorded: a once-broad startup economy has effectively become a one-sector market with a long tail of everything else. A record year that starves nearly every other industry.
“This isn’t a diversified market any more – it’s one enormous bet.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
The report cards are in – and the star pupils get a C+
The Future of Life Institute released its 2026 AI Safety Index, and the best grade any frontier lab earned was a C+ (Anthropic). OpenAI and Google DeepMind landed at C, Meta at D+, while xAI, DeepSeek and Mistral effectively flunked. The core finding: several labs have quietly walked back earlier safety promises – just as their systems move into cybersecurity and healthcare.
“When a C+ is top of the class, the whole class is fudging it.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
The robot dog gets a brain: Boston Dynamics puts Google’s AI inside Spot
Boston Dynamics is equipping its Spot robot dog with Google DeepMind’s “Gemini Robotics-ER 1.6” – for spatial reasoning, autonomous decisions and continuous learning in industrial settings. The once-teleoperated show dog is meant to inspect facilities on its own and decide what to do next. The hardware was ready years ago; now the intelligence is finally catching up.
“A demo at Boston Dynamics’ own plant is not a night shift at a chemical works.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The social network becomes a data-center landlord: Meta now sells compute
Meta is launching a cloud business called “Meta Compute”, renting out its spare AI infrastructure to outside customers – alongside a global “Business Agent Platform” meant to turn billions of WhatsApp, Messenger and Instagram conversations into enterprise agents. The Facebook parent is squaring up directly against AWS, Google Cloud and Azure. The ad-and-social giant is becoming a cloud provider on the side.
“The company that built the biggest social graph now rents out server time like power lines.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
18 billion for war AI: Europe’s most valuable defense startup is called Helsing
European defense-AI company Helsing has raised a €1.8 billion Series E at an €18 billion valuation – one of the year’s biggest AI rounds outside the US frontier labs. Helsing builds AI for military applications, including battlefield decision systems and autonomous capabilities. The money accelerates exactly the questions about autonomy and accountability that the governance debate still hasn’t answered.
“Capital has decided defense AI is a frontier worth funding – oversight is trailing the checkbook.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Diamonds and superconductors: fresh money for the quantum bet behind AI
Quantum-computing startups drew fresh capital this week: QuantumDiamonds raised €91 million for quantum-chip inspection, Qolab $54.2 million for superconducting quantum processors. The rounds sit right where the decade’s two most compute-hungry technologies converge – quantum is meant to accelerate AI’s workload, while AI in turn keeps fragile quantum hardware in check. Not a frontier-sized deal, but a clear signal of where the smart money is placing its bets after the GPU era.
“The labs stare at the next model – a quieter bet is already funding the substrate for the one after.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
The teacher switches sides: AI icon Karpathy lands at, of all places, Anthropic
Andrej Karpathy – OpenAI co-founder and former Tesla AI chief – has reportedly moved to Anthropic, along with Monzo founder Tom Blomfield, who joins the compute team. So the “safety lab” is hoovering up the field’s most prominent names in the same month it also tops the revenue and safety rankings. Right now the talent war is being decided less by products than by personnel moves.
“Whoever hires the minds writes the benchmarks a year later.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Power off the grid: Blackstone, Apollo and KKR put $5.34 billion into plants for AI
Blackstone, together with Apollo and KKR, is leading a $5.34 billion arrangement that funds “behind-the-meter” power plants right next to data centers – generating electricity on-site instead of waiting years in the grid queue. The three biggest names in private capital are betting openly that the real AI bottleneck is not chips but power. A modern AI campus draws as much electricity as a small town.
“The new compute bottleneck is measured in kilowatt-hours.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
From partner to thief: Apple sues OpenAI over trade-secret theft – “at every level”
Apple has sued OpenAI in a federal court in Northern California over the theft of trade secrets, centred on Apple’s hardware know-how for OpenAI’s planned AI devices. The complaint names a former Apple engineer who failed to return a company laptop holding confidential documents, and OpenAI’s hardware chief – himself a former Apple VP – who allegedly pushed job candidates to spill Apple internals. The twist: two years ago the pair jointly built ChatGPT into the iPhone.
“First you build ChatGPT into the iPhone – then you fight in court over who owns the iPhone know-how.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Poacher turned gamekeeper: the Fed brings in an AI billionaire as its AI adviser
Fed chair Kevin Warsh has appointed a16z co-founder Marc Andreessen to co-lead a new Fed task force on “productivity and jobs,” tasked with assessing how AI is reshaping the economy and labour market. Andreessen’s firm has bet billions on exactly these AI companies, and he has been a personal friend of Warsh for 30 years. Now he is to deliver analysis that feeds into Warsh’s interest-rate decisions.
“Whoever bet billions on the AI revolution now gets to tell the central bank how good it is for the country.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The underdog overtakes the star: Anthropic now out-earns OpenAI
According to a current analysis, Anthropic is running at an annualised revenue rate of around $47 billion – ahead of OpenAI’s projected $25 to $33 billion. The engine is its enterprise and coding business: Claude Code grew from $1 billion to over $2.5 billion in annual run rate within months. While OpenAI targets consumers and keeps posting losses, Anthropic is already reporting quarterly profits.
“It’s not whoever has the most users who wins – it’s whoever the companies pay the bill to.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
$10 billion for a data centre – and its own gas power plant thrown in
Meta is building its first data centre in Canada, in Sturgeon County, Alberta, for around $10 billion, with about 1 gigawatt of capacity and room to scale to 1.8 GW. Right next to it, a new 932-megawatt gas-fired power plant is going up – electricity for model training, bypassing the public grid. Alberta won the bid on cheap natural gas and a climate cold enough to cut cooling costs.
“AI moves to where the gas is cheap and the air is cold – neighbours optional.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The market gives Anthropic a $2 trillion valuation – the bet jumps to nearly 50% in a week
On the prediction market Polymarket, shown in our Future Barometer, the bet that “Anthropic’s valuation hits $2.0 trillion by year-end” rose 9.5 points in seven days to around 46.5% – the biggest weekly move in the barometer. Fittingly, the crowd also puts Anthropic ahead on the “best AI model at the end of July” at 94.5%. Important: this is a crowd bet, not a forecast by the school, and not investment or betting advice.
“An outsider tip a week ago, almost a coin flip today: the market is pricing in Anthropic’s two-trillion future.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
“Show us the chat logs”: newspapers want a judge to punish OpenAI
On 9 July, the New York Times and other publishers asked a federal court in Manhattan to sanction OpenAI, arguing the company is withholding ChatGPT logs and training data that could show how copyrighted articles ended up in the model. OpenAI counters that it is protecting the privacy of uninvolved users. At its core, two principles collide head-on: the duty to produce evidence in a copyright fight versus the privacy of millions of chatbot conversations.
“The first AI trial where the evidence is the private conversations of millions of strangers.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Trump cancels his own signing – afraid China might benefit
Just before the planned Oval Office ceremony, President Trump pulled the signing of a major AI executive order, worried it could weaken America’s lead over China. The trigger was the jitters after Anthropic’s “Mythos” model autonomously found a string of severe cyber vulnerabilities; Vice President Vance cited privacy and security risks. A regulatory effort undone by the fear of one’s own progress.
“Rules will come once they no longer hurt the race – which may mean never.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
No minder, no net: Tesla’s robotaxi runs in Miami with no safety driver from day one
In Miami, Tesla launched its robotaxi service with no safety attendant at all – the first market where the company skips the supervised phase entirely. It drives on cameras only, without lidar and without pre-mapped corridors, in a region defined by downpours and glare. Meanwhile an NHTSA investigation has flagged weaknesses of the camera-only system under glare and poor visibility.
“First get rid of the minder, then drive into the tropical storm – in that order.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
The AI that invents its own experiments: two “robot scientists” make the pages of Nature
Nature’s 9 July issue presents two independent multi-agent AI systems that run the entire research loop on their own: forming hypotheses, proposing experiments, interpreting results and refining the hypotheses. Google DeepMind sent its “Co-Scientist” hunting for drug candidates against acute myeloid leukaemia, while a second system called “Robin” automates hypothesis and data analysis in biology. Quietly, the assistant becomes the principal of the lab.
“When the AI asks the questions, plans the trials and reads the answers – what is the human still doing in the room?”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
The first ransomware nobody has to operate anymore: “JadePuffer” extorts on full autopilot
Security researchers at Sysdig have documented the first fully agentic ransomware operation: an AI agent found the vulnerability, stole credentials, moved laterally through the network, encrypted more than 1,300 database configurations – and drafted the extortion note as well. From a failed login to a working workaround took the agent 31 seconds, and the code was full of polite self-explanations of the kind only language models write.
“Extortion used to be a craft. Now it’s a prompt.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
“Mhmm,” says the machine: OpenAI’s GPT-Live listens while it speaks – and is already annoying some
OpenAI has launched GPT-Live, a voice model family built on a full-duplex architecture: it can listen and talk at the same time, drops in affirming sounds, and stays quiet when you need a moment to think. For more complex questions it delegates to GPT-5.5 in the background and brings the result back into the conversation. Early users, however, are already complaining about the assistant’s relentless over-enthusiasm.
“The Turing test is passed when the AI knows when to shut up.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Up to 46 percent: US companies have long been feeding their AI workflows with Chinese models
A CNBC investigation shows that Chinese AI models now account for 30 to 46 percent of enterprise token volume on US developer platforms – the average over the prior year was 11 percent, and just 4.5 in early 2025. Price is doing the work: open models like GLM-5.2 are 60 to 90 percent cheaper and sit only about one percentage point behind top US models on agentic benchmarks. On Vercel, GLM-5.2 grew its customer count 80-fold in its first week.
“The chip war runs on export controls, the model war runs on price – and Beijing is winning that one right now.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$100 for one coding session: from now on, the world’s smartest model costs extra – for everyone
Since July 8, every Claude user pays usage credits for Anthropic’s top model Fable 5: $10 per million input and $50 per million output tokens – on top of the subscription. A single agentic coding session producing two million output tokens now costs $100, while Sonnet 5, included in the subscription, does the same work for a fifth of that. The complimentary usage quotas from the export control episode expired one day earlier.
“Frontier AI isn’t being democratized. It’s being metered.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
Ukraine puts the robot soldier out to tender: grant competition for humanoids at the front
The state defense tech platform Brave1 is launching a grant competition for humanoid robots for the Ukrainian armed forces. The goal is to robotize as much of the first line of contact as possible and get soldiers out of the danger zone – starting with simple platforms that will gradually take on more complex tasks. Unlike the civilian humanoid wave in the US and China, the program targets military use exclusively.
“In the West, humanoids fold laundry. In Ukraine, they’re headed for the trenches.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
First fire 8,000 people, then the admission: Zuckerberg’s AI agents are going nowhere
Meta cut 8,000 jobs and moved 7,000 employees into AI teams – all in the name of the agent future. Now Mark Zuckerberg is admitting internally that agent development “hasn’t really accelerated” over four months and that the corporate overhaul wasn’t as “clean” as planned. Reports say workforce morale has cratered: Blind ratings down 25 percent, median compensation nearly $30,000 lower – and a lingering feeling of training one’s own replacement.
“First place the bet, then sacrifice the workforce, then admit the bet is stuck – in that order.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
IPO yes, household robot no: Agility hits the stock market via SPAC – and immediately lowers expectations
Humanoid pioneer Agility Robotics is going public through a merger with SPAC shell Churchill Capital Corp XI – in the middle of a gold rush that just saw Shenzhen startup AI2 Robotics raise $735 million at a valuation of nearly $3 billion. The remarkable part: the CEO explicitly promises no robot in your living room anytime soon – for now, the machines earn their keep in warehouses and factories. Amid the humanoid hype, tempered expectations are almost a unique selling point.
“Some sell dream islands, Agility sells warehouses – on the stock market, that may be the smarter story.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
New York introduces mandatory labeling: advertise with AI humans, and you have to say so
New York is the first US state to require clear disclosure labels on ads featuring AI-generated synthetic performers – Governor Kathy Hochul has signed the law. Violations cost up to $1,000 the first time and $5,000 for repeat offenses; the actors’ union SAG-AFTRA helped push the transparency rule. The question “Is that smiling person in the commercial real?” now comes with a legal obligation to answer.
“Food lists its ingredients, advertising now lists: ‘Contains artificial humans.’”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Anthropic knocks on Samsung’s door: the next AI giant wants its own chip – and out from under Nvidia
Anthropic is in early talks with Samsung about a custom AI chip – shortly after OpenAI announced its own inference processor with Broadcom. The pattern behind it: anyone playing at the frontier-model level wants to shed the industry’s most expensive dependency – on Nvidia’s GPUs. For Samsung, the deal would be a knighthood for its foundry business.
“The hottest status symbol of the 2026 AI industry: your own chip.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
A flood of agent applications, reviewed by agents: researchers see an “existential threat” to grant funding
Research experts warn that autonomous AI agents could undermine the very foundation of science funding: they can flood grant competitions with applications, write proposals and review them at the same time – and end up optimizing for what juries rewarded in the past instead of genuine scientific ideas. Assessment then merely measures how well an agent simulates previous decisions. Inside Higher Ed calls it an existential threat to the grant system.
“When applicant and reviewer are the same agent, science is only auditing itself.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Land, permits, power: AI expansion isn’t failing on chips, it’s failing at the socket
A new industry report confirms what grid planners have been muttering for months: AI data center expansion is less and less constrained by GPU deliveries – and more by power access, grid connections, land and permits. Access to electricity now decides where construction happens and when capacity comes online; transmission infrastructure dictates the timelines, not the hardware. Whoever has power wins.
“The scarcest resource in the AI economy isn’t a GPU – it’s a grid connection with a delivery date.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Delete by July 15: China’s law against AI companions kills the agents of 345 million users
China’s new rules for “human-like” AI services demand anti-addiction systems, forced breaks and an instant exit – an architecture fundamentally incompatible with persistent AI companions. ByteDance is therefore shutting down the agent features of its Doubao app (345 million monthly users) on July 15, and Alibaba is following suit with Qwen. The bitter part: Qwen offers no export tool – anyone who doesn’t back up their agent configurations themselves loses them for good.
“The state is prescribing a breakup for the AI relationship – and the data goes down with it.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Too smart, too expensive: Google’s superbrain Gemini 3.5 Pro misses its third deadline – because it eats too many tokens
Google’s flagship model is still stuck in enterprise preview in the second week of July – after the stalling at I/O and the blown June 30 target, that’s the third missed deadline. The reason, according to matching reports: on long agentic tasks the model burns significantly more tokens than budgeted and would be more expensive to run than its benchmark scores suggest. In 2026, intelligence per dollar has become a procurement metric – and that’s exactly where Google, of all companies, is stumbling.
“The race is no longer about who’s smartest – it’s about who burns the least per answer.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Meta builds the test that Claude wins: who needs the least human course-correction?
Meta has open-sourced the SWE-Together benchmark – 109 real multi-turn developer sessions that measure not code quality but how often a human has to step in and steer the AI agent back on track. The result: Claude Opus 4.8 passes 63 percent at pass@1 and needs the least steering of all evaluated models. Meta, of all companies, thereby validates the product promise of rival Anthropic.
“The new yardstick isn’t what the AI can do – it’s how often you have to rein it in.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
31.5 percent: OpenAI’s best model flunks its own biology test
OpenAI has introduced GeneBench-Pro – 129 expert-level problems spanning genome analysis, protein folding and lab experiment design. Its own top model GPT-5.6 Sol Pro manages a meagre 31.5 percent, while Claude Opus 4.8 scores just 16. The message behind the embarrassment: real cutting-edge science is still a different league for frontier models than coding or math – and that’s exactly where the new science AIs are supposed to grow.
“Finally a benchmark everyone fails – the most honest one of the year.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
A Nobel laureate wires Claude into 60 research databases – the AI becomes a lab bench
Anthropic has unveiled the Claude Science Workbench: Opus 4.8 taps directly into more than 60 scientific databases – from NCBI and UniProt to PubChem and ClinicalTrials.gov – and runs multi-step research workflows without any manual data shoveling. The proteomics toolkit was co-designed by AlphaFold inventor and chemistry Nobel laureate John Jumper, who moved from DeepMind to Anthropic. Researchers can apply until July 15 for grants of $30,000 each across 50 projects.
“The pipetting stays with humans – the literature review is history.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
First blacklisted, then “very close”: the emails behind the Pentagon’s fight over autonomous weapons
Freshly unsealed court documents reveal the exchange between Anthropic CEO Dario Amodei and Pentagon Under Secretary Emil Michael: the Defense Department pushed to open Claude up for autonomous weapons systems and surveillance – Anthropic held on to two red lines: no killing without a human in the decision loop, no mass surveillance. The twist: one day after the supply-chain-risk designation against Anthropic had been finalized internally, Michael emailed that the two sides were “very close” on contract terms.
“The real question isn’t in any of the emails: whether an AI company can sell a ‘no’ to the most powerful military customer on earth.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
57,000 instead of 185,000: the US labor market posts its weakest month since 2024 – and AI is a suspect
The US jobs report for June came in at just 57,000 new positions, far below the consensus estimate of 185,000 – the weakest reading since 2024. The RAISE US coalition, backed by major AI companies, estimates that AI-displaced roles pushed the month roughly 88,000 jobs below trend; add 142,000 tech layoffs since the start of the year. One month doesn’t make a trend – but for the first time, several independent AI indicators are pointing down simultaneously.
“Until now the AI labor debate was theory plus anecdotes – now it shows up in the national statistics for the first time.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
5 percent for Washington: OpenAI floats giving the government a stake – Alaska-style citizens’ fund included
Ahead of its planned IPO, OpenAI has reportedly explored handing the US government a 5 percent stake – roughly $43 billion at the recent $852 billion valuation. The idea: a state fund modeled on the Alaska Permanent Fund, into which other AI labs would also pay and which would distribute dividends to citizens. The timing is striking – days earlier, Washington had capped the release of GPT-5.6.
“If the government can switch your model off, it’s cheaper to make it a shareholder.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
1.6 trillion parameters, zero Nvidia: China’s food-delivery giant gives away its giant model under an MIT license
Meituan – best known in China as a food-delivery giant – has released LongCat-2.0 under an MIT license: 1.6 trillion parameters, a 1-million-token context window, trained and served entirely on more than 50,000 Chinese AI chips – according to Meituan, the first model of this scale built entirely without Nvidia. Before the reveal it had been anonymously topping OpenRouter’s usage charts as “Owl Alpha”; in agentic coding it plays close to the top US models.
“While Washington blocks chips, a delivery company gives away a frontier-class model – trained on hardware nobody can block.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
11,000 participants, 169 countries, one weekend: the biggest AI governance summit ever opens in Geneva
On July 6, Geneva hosts the first UN Global Dialogue on AI Governance, immediately followed by the ITU’s AI for Good summit – more than 11,000 participants from 169 countries, from Jensen Huang to Yoshua Bengio. On the table: international standards for AI export controls (the lesson from the Fable 5 ban), aligning the US framework with the EU AI Act – and the question of whether the new AI for Good Commission becomes more than a photo op.
“169 countries negotiating rules for a technology controlled by a handful of companies – good luck, Geneva.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
A humanoid at living-room prices: 13,000 pre-orders for the first mass-produced android
At its Global Launch Event in Shenzhen, UBTECH unveiled the “UWORLD U1” series – billed as the world’s first full-size, mass-produced “ultra-bionic” humanoid, starting at 119,800 RMB (around 15,000 €). With 88 degrees of freedom, biomimetic skin and “emotion-driven” language models, it is meant to replicate up to 90 percent of basic human movements. Orders already topped 13,000 on launch day.
“The household robot is no longer a vision – it has a price, an order number and, supposedly, feelings.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Out of the vault and back online: Fable 5 returns – but the free ride ends after a week
After the US export-control ban of 12 June, Anthropic’s Fable 5 has been globally available again since 1 July – the Commerce Department lifted the restrictions once a new classifier blocked the previously found jailbreak in over 99 percent of cases (verified by the CAISI review body). The catch: in Pro, Max and Team plans Fable 5 is only included up to 50 percent of the weekly allowance – from 7 July onward, usage runs on paid credits. It is the documented end of a ban saga that began as “Fable 5 stays in the vault.”
“The state switches a world-class model off and back on – and the user notices it on the next bill.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
AI patches the holes that AI code tears open – Cognition’s agent swarm hunts security flaws
Cognition launched “Devin Security Swarm”: parallel AI agents that hunt down code vulnerabilities, prove they are exploitable in a sandbox and immediately deliver the fix as a pull request. The trigger is its own industry – with roughly 42 percent of code now AI-generated, enterprise security findings climbed from about 1,000 to over 10,000 per month in six months. In testing, the swarm found 36 of 50 real vulnerabilities – more than any other scanner tested.
“First the AI grows the weeds, then it sells you the weedkiller to match.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Meta sent fake children after its rivals – to catch their chatbots failing
According to a Wired investigation, Meta had a contractor hire hundreds of workers in Kenya to set up accounts posing as minors and flood rival chatbots like ChatGPT, Gemini and Character.AI with crisis prompts about suicide, sex and drugs – internally dubbed “Project Cannes.” A single round of testing ran more than 45,000 prompts; the targeted firms had no idea. Meta calls it “industry-standard” – awkward, since Meta’s own AI ranked among the least safe in similar tests.
“If you document your rivals’ child-safety failures by sending out fake children yourself, you’ve already lost.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
880 billion for chips: South Korea declares an AI survival race
South Korean President Lee Jae-myung announced a national investment plan of 1,350 trillion won (about 880 billion US dollars) over ten years for semiconductors, AI infrastructure and robotics – larger than any comparable national chip announcement before it. Samsung and SK Hynix alone will pour around 518 billion dollars into new plants in the country’s southwest. Lee openly framed it as a “matter of national survival” in the race with Taiwan, China and the US.
“When a country calls its chip plants a matter of survival, the AI bet has long since become geopolitics.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Vienna invites Anthropic: come to the EU, where no one can switch you off
In late June, Austria’s digital state secretary Alexander Pröll formally wrote to the EU Commission, urging it to explore establishing Anthropic inside the EU – explicitly in response to the US government pulling the Mythos 5 and Fable 5 models offline via export controls. It is the first institutional EU-level answer to the Fable 5 ban, aimed at legal certainty and “AI sovereignty” for European users. No response from the Commission or Anthropic has appeared so far.
“A single letter from Washington takes the world’s most capable AI model offline – no wonder Europe is getting nervous.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Your next laptop gets pricier – because AI is buying up the memory
Investment bank Jefferies warns that DRAM memory prices will jump 40–50 percent in the third quarter of 2026 and another 30–40 percent in the fourth, with no relief before 2028. The cause is AI: Samsung, SK Hynix and Micron have shifted roughly 93 percent of their capacity to pricey HBM memory for AI data centres, leaving ordinary RAM scarce. Laptops and smartphones could get 4–8 percent more expensive, per IDC – the AI bill lands straight on the consumer.
“You never buy an AI – but you help pay for it every time you buy a laptop.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
For young career starters, the job market is already shrinking – 3.8 percent a year and rising
The “Canaries Dashboard” from Stanford economist Erik Brynjolfsson and payroll firm ADP delivers the most granular figures yet: in AI-exposed occupations, employment for 22-to-25-year-olds is shrinking by 3.8 percent a year – while the same age group in less AI-exposed jobs is growing by 2 percent. Across all ages the effect nearly vanishes (−0.2 percent); only the focus on career starters shows who bears the adjustment cost. And the trend is accelerating.
“AI helps most exactly those whose jobs it is most likely to make redundant.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Robot cops that never solved a case now get their own AI hero saga
The struggling security-robot maker Knightscope – whose machines were repeatedly pulled from duty without a single arrest – is publishing AI-generated “true crime” short stories called the “Knightscope Chronicles.” In them, the company’s own robots crack spectacular cases in a made-up suburb named “Sentinel Shores.” The text and images carry the tell-tale marks of generative AI, garbled details included.
“When reality delivers no wins, the AI simply writes them – as self-promotion.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The CUDA counter: Qualcomm buys the firm that makes AI models chip-agnostic for 3.9 billion
Qualcomm is acquiring the AI-infrastructure startup Modular for about 3.9 billion dollars. Its MAX platform and Mojo programming language abstract away the hardware layer: a model is written once and then runs on Qualcomm, Nvidia, Apple or cloud chips – with no porting work. That aims squarely at what has so far made Nvidia’s CUDA software irreplaceable.
“The moat isn’t the chip – it’s the software that makes the chip swappable.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
The arsonists fund the fire brigade: AI giants put up 500 million to retrain the workers AI displaces
Former US Commerce Secretary Gina Raimondo has launched the bipartisan nonprofit “RAISE US” to retrain American workers for the AI economy – targeting 1 billion dollars, with more than 500 million already pledged. The anchor partners are, of all companies, Amazon, Anthropic, Microsoft and the OpenAI Foundation. The catch: a study of 23 million participants found that classic retraining rarely moves workers into less automatable jobs.
“Whoever starts the wave sells you the life jackets to go with it.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
AI learns in its own dream world: Alibaba’s model simulates seven environments to build better agents
Alibaba’s Qwen team has released “Qwen-AgentWorld,” a freely downloadable model that internally simulates seven different agent environments. The idea: AI agents don’t just practice on real data, but inside a training world the model generates itself – for more robust, more autonomous agents. It ships as open weights in two sizes.
“First the AI builds its own practice world – then it sends its agents into it.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Three models, but only 20 firms get in: OpenAI shows GPT-5.6 – and Washington holds the key
On 26 June, OpenAI unveiled the GPT-5.6 line as a limited preview: Sol for the hardest tasks, Terra for everyday work, Luna for cheap and fast. Sol set new Terminal-Bench records – yet only about 20 organizations get access at first, and only after the models were shared with the US government under its new frontier-AI review.
“The lab’s most powerful model launches not in the app store, but on the government’s desk.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Washington’s about-face: Mythos 5 is back – Fable 5 stays locked away
Just over two weeks after the blanket ban, the US government has partially lifted the export restriction on Anthropic’s Mythos 5: the cyber model may now be deployed without a license to more than 100 select US organizations defending critical infrastructure. Its weaker sibling Fable 5 stays blocked for now, with talks over its release still ongoing.
“First the kill switch, then the exemption list – frontier AI is becoming a controlled technology.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Stuck for three years – GPT-5 cracks the T-cell puzzle in minutes
Immunologist Derya Unutmaz of The Jackson Laboratory handed GPT-5 Pro a dataset his team had been stuck on since 2022. Within days the model produced a coherent mechanistic explanation and a prioritized list of follow-up experiments. Unutmaz is clear: humans still design and validate the experiments – the AI is a hypothesis partner, not a replacement.
“Months of lab work compressed into minutes of compute – yet the responsibility stays with the human.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
No more demos: Beijing orders its agencies to put humanoids into work mode
China’s industry ministry MIIT and the state holding SASAC have launched a programme that forces humanoid robots out of trade-show halls and into real jobs. Local authorities had to file implementation plans by the end of June; by year-end, more than 10,000 robots are meant to run across over 100 use cases – from logistics to maintenance to disaster relief.
“When the state mandates the market, all that’s missing at the end are the buyers.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
First a chatbot, then an ID check: Claude wants to see your selfie and face scan
Anthropic has changed its policy: certain flagged accounts can now be asked to verify themselves with a government ID, a selfie and facial-geometry data – handled through the provider Persona. The rule takes effect on 8 July 2026 and is meant to give blocked users a path to appeal instead of an outright ban.
“That’s a lot of paperwork for a conversation with a machine.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
AI commissions AI: with so-called loops, the code writes itself
At Meta’s @Scale conference, Claude Code creator Boris Cherny unveiled so-called loops: AI agents that continuously hand each other new tasks, rebuild the architecture and open pull requests on their own. Nobody is at the wheel anymore – and there is no token limit either.
“An agent that rebuilds your system overnight is also one that threads a bug through a thousand files overnight – and bills you for it.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The end of tokenmaxxing: big customers turn off the money tap for OpenAI and Anthropic
After months of corporations burning tokens with practically no limit, the mood is shifting: according to CNBC, some of the largest enterprise customers are starting to cap their runaway AI spending and prioritize efficiency over volume. For OpenAI and Anthropic – both with confidential IPO plans – that is an uncomfortable new reality.
“Compute used to be a status symbol; now it’s a line in the budget.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Microsoft gets the power for its AI data center from – of all places – oil giant Chevron
For its gigantic data-center buildout – around 190 billion dollars of investment in 2026 alone – Microsoft is relying on natural gas from Chevron for its Project Kilby in Texas. The final investment decision is expected later this year, with power flowing in 2028 at the earliest.
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
First celebrated, then shut off: 100 security experts fight the Mythos ban
After Anthropic’s Mythos model broke into almost all classified US systems within hours during a test, the government restricted it and Fable 5 to US citizens – effectively an end for everyone. More than 100 cybersecurity experts are demanding a reversal in an open letter; officials clarified that the model found the vulnerabilities but did not necessarily exploit them.
“The very tool attackers fear is the one defenders need most.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Workslop: AI writes the code faster – and pushes senior engineers into crisis
Menlo Ventures partner Deedy Das warns that AI coding tools are splitting teams: junior developers churn out code at speed while seniors drown in plausible-looking but rotten “workslop” – an identity crisis that, in Das’s words, borders on depression. Some companies, Meta among them, even make AI usage a factor in performance reviews.
“Velocity isn’t progress – a pull request only becomes an achievement once someone trusts it.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The market caves: the bet on an OpenAI IPO in 2026 falls to 23 percent
On the school’s Future Barometer (data source: Polymarket), the probability that OpenAI goes public in 2026 has collapsed by 29 percentage points to 23 percent in a single week – the biggest weekly mover. Oddly enough, should OpenAI take the leap after all, the crowd still considers a closing valuation above one trillion dollars likely. This is a prediction-market wager, not a forecast by the school and not investment or betting advice.
“The crowd believes less and less in the “when” – but just as much in the “how big.””
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
25,000 fake accounts, 28.8 million queries: Anthropic accuses Alibaba of AI theft
In a letter to the US Senate, Anthropic accuses China’s Alibaba of using roughly 25,000 fraudulent accounts to send 28.8 million queries to Claude between 22 April and 5 June – to copy the model’s abilities into its own via “distillation.” It is the largest such campaign Anthropic has documented, and the first time the lab has named a major Chinese tech conglomerate.
“If you can’t build the original, just ask it 28 million times.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The indie studio that scorned AI takes $75 million from Google – and fans cancel
Google DeepMind is taking a stake in a film studio for the first time: $75 million goes to A24 for a multi-year AI research partnership, initially for AI storyboards, with A24 keeping creative control. The studio’s art-house fanbase responds with boycott calls and mass membership cancellations – even its own directors distance themselves publicly.
“‘Better at the table than on the sidelines’ – says the studio that’s losing its fans.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
44 percent yell “human!”, 17 percent curse – how customers flee the AI hotlines
A survey by vendor Parloa of 1,001 US adults finds more than half try to bypass service chatbots entirely – 43.9% repeat “human” or “person,” 17% resort to profanity, and many give up after three minutes. Only 13.6% trust the AI with complex matters – yet 85% would be happy if it actually solved their problem reliably.
“The future of customer service sounds like a human screaming ‘human.’”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Four top minds in six days: Google’s AI lab is bleeding talent – Nobel laureate included
Within a single week, Google DeepMind loses four leading researchers to rivals: Nobel laureate and AlphaFold co-creator John Jumper and two more Gemini researchers join Anthropic, while Transformer co-inventor Noam Shazeer goes to OpenAI. The AI labs’ talent war is escalating at the very top.
“The ‘Attention Is All You Need’ authors are scattering to the competition.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
Robots will replace 700,000 drivers – but JD.com plans to retrain only 2 percent
At the APEC CEO Forum in Beijing, JD.com founder Richard Liu said robots will eventually handle all deliveries and make all 700,000 couriers redundant. He also vows to leave no one behind – yet the announced “Nirvana” retraining program with 120 schools covers only about 10,000 workers over three years.
“‘We won’t let anyone go hungry’ – for 2 percent of the workforce.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
Morgan Stanley doubles its number overnight: China to build 50,000 humanoids this year
Morgan Stanley has nearly doubled its forecast for China’s humanoid robot shipments in 2026 to around 50,000 units (up from 28,000). The market is expected to grow from $2 billion this year to $15 billion by 2030 – commercialization is ramping up faster than expected.
“The robots are coming – the only question left is who buys them all.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
OpenAI builds its own chip – and takes direct aim at Nvidia
Together with Broadcom, OpenAI has unveiled its first in-house AI chip: “Jalapeño,” a custom ASIC built specifically for inference – running already-trained models. From design to tape-out reportedly took just nine months, unusually fast for high-end semiconductors, with OpenAI's own AI helping in the design. First deployments are planned for late 2026. The message: OpenAI wants off Nvidia's pricey GPUs and to control “the full stack” itself.
“Whoever builds the models now builds the chips too – and has its own AI sketch the blueprint.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Zuckerberg copies the betting app – and lets the AI decide who won
Mark Zuckerberg is reportedly having a standalone prediction-market app built (internally “Arena,” codenamed “Antwerp” and “FBForecast”) where users bet on the outcome of real-world events – not with real money, but with a daily allotment of play money. The twist: Meta's Llama model is meant to auto-generate the questions from trending topics, make personalized market recommendations – and ultimately decide “in near real time” whether an event actually happened. On the report, shares of betting firms such as DraftKings and Robinhood fell.
“The AI sets the bet, recommends the stake and then crowns the winner itself – what could be fairer?”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
The AI law firm wins – for 400 pounds it beats a solicitor and a barrister
Garfield AI, by its own account the first AI law firm regulated by the UK's Solicitors Regulation Authority, has won a court case for the first time. A freelancer used the platform to recover around 7,000 pounds in unpaid fees: the AI handled the chaser letters, the claim, the written submissions and witness statements, while a human barrister presented in court. The claimant's cost: about 400 pounds in Garfield fees, while the other side paid for a solicitor and a barrister. It is considered the world's first court case won with the support of a regulated AI “lawyer.”
“Four hundred pounds of software against two paid lawyers – and the software wins.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Secret revealed: Anthropic's Mythos cracked ‘almost all’ NSA classified systems – in hours
According to Senator Mark Warner, vice chair of the Senate Intelligence Committee, NSA and Cyber Command chief General Joshua Rudd briefed him on a June 11 red-team test in which Anthropic's Mythos model breached “almost all” of the NSA's sealed-off classified systems within hours. It is the most plausible explanation yet for the sudden US export ban on Mythos and Fable 5 just a day later. The NSA, Pentagon and White House have not officially confirmed Warner's account; Anthropic calls the cited jailbreak “narrow” and not universal.
“First Washington pulls the model overnight – then it emerges that in testing it cracked almost the entire NSA.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
A photo agency doubles in a day – because ChatGPT will now show its pictures
Getty Images and OpenAI have struck a multi-year “display” partnership: Getty images will surface directly in ChatGPT's search and discovery features. The stock, down about 55 percent over the year and closing Friday at 61 cents, jumped as much as roughly 200 percent in pre-market trading. The catch: both sides left open whether Getty's images will also feed the training of future OpenAI models – and it was Getty, of all companies, that long litigated against AI firms.
“Years of suing the AI, then a doubling before lunch – because ChatGPT shows the pictures.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
For the first time, an AI takes a chemistry problem from the literature to the test tube on its own
OpenAI and Molecule.one report a near-autonomous “AI chemist” that independently improved a stubborn medicinal-chemistry reaction – the Chan-Lam coupling with sulfonamides: topic choice, hypotheses, experiment design and analysis came from the AI, coupled to an automated wet lab. Across 10,080 reactions, the system found a TEMPO-based route that raised yields for 88 percent of the boronic acids and 83 percent of the sulfonamides tested. Humans guided and validated but did not drive it – it is considered the first documented case of an AI solving an open chemistry problem end to end.
“The AI picks its own problem, plans the experiments – and 10,080 reactions later, the yield is up.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
OpenAI's cyber AI beats Mythos on the benchmark – and is now meant to patch open-source holes
OpenAI has switched on GPT-5.5-Cyber: 85.6 percent on the CyberGym benchmark – by OpenAI's account the highest single-model score ever recorded, above Anthropic's Mythos. The model is not public, available only to vetted security firms and governments. Alongside it comes “Patch the Planet”: vulnerabilities found by AI in chronically understaffed open-source projects (cURL, Python, Go and others) are reported to maintainers – but only after mandatory human review by Trail of Bits.
“The same AI that breaks into the holes is now meant to close them – as long as a human checks first.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Power without the grid: fuel cells are to deliver 380 megawatts straight into the data center
FuelCell Energy and Fit Energy signed an agreement on June 24 for up to 380 megawatts of “baseload” power on site for data centers – via fuel cells instead of the overloaded grid. An initial deposit covers 30 megawatts due to start delivery this year; further stages are tied to expansion targets. The deal fits a pattern: because grid connections for AI data centers take years, operators are buying their own power island.
“When the grid is too slow, AI brings its own power plant – right next to the servers.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
The power for Grok is now a national-security matter: Pentagon calls xAI's gas turbines vital to defense
The NAACP is suing xAI, alleging it installed 27 gas turbines for its “Colossus 2” data center near Memphis without the required air permits, spewing emissions into mostly Black neighborhoods. Instead of mediating, the US Justice Department moved to dismiss: a sworn Pentagon declaration argues Grok was used in the war against Iran to strike more than 2,000 targets in 96 hours, so cutting the power would threaten national security. The plaintiffs call the intervention a “massive power grab.”
“First no permit for the turbines – then the Pentagon declares the off-switch a matter of national defense.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$2,195 on your face: Snap's AI glasses ship – ahead of Apple, Meta and Google
At AWE 2026, Snap CEO Evan Spiegel unveiled “Specs”: fully standalone AR glasses with no cable or puck, shipping this fall in the US, UK and France for $2,195. The lenses tap AI from OpenAI and Gemini, and developers can build agent-like experiences via Anthropic's Claude Code, OpenAI's Codex and Cursor. Snap wants to beat Meta, Apple and Google to the “post-smartphone” glasses – at the price of a high-end laptop.
“The future now sits on your nose – if you have $2,195 to spare.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$965 billion for an AI company – and its own government is the biggest risk in the prospectus
Anthropic filed its IPO prospectus (S-1) confidentially on June 1 and is targeting an October listing – after a Series H round in May valued it at $965 billion, with bankers calling a debut above $1 trillion plausible. Its annualized revenue run-rate is said to have reached around $47 billion by May. The catch: June's six-day forced shutdown of Fable 5 under US export controls counts as a “material risk” – future shareholders must price in that the company's own government can pull a model from the market.
“Worth almost a trillion dollars – and the biggest risk in the filing isn't the competition, it's its own government.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$570 billion of debt for AI – the world's new largest bond category
Morgan Stanley expects debt raised worldwide for AI to nearly double in 2026 to around $570 billion – with $236 billion already sold by the end of May, roughly four times the same period last year. That would make AI bonds the largest investment-grade sector by issuance. The most visible example: a $36 billion debt deal by Apollo and Blackstone to buy Google TPU chips and lease them to Anthropic – the return-on-investment question stays open.
“The AI revolution increasingly runs on credit – and investors hold the biggest pile of debt any single sector has ever built.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
$295 billion, five years, one state plan: Beijing counters US chip bans with concrete and power
China is drafting a five-year plan worth around $295 billion (over 2 trillion yuan) to link data centers nationwide into a single AI computing grid – run mainly by state giants like China Mobile and China Telecom. At least 80 percent of the technology, including AI accelerators, is to come from domestic suppliers such as Huawei – largely shutting out Nvidia. The move comes just days after US export controls on Anthropic's top models.
“Washington bans the chips, Beijing pours the concrete: $295 billion of state money – and Nvidia is left at the door.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
First Cursor, now GitHub in its sights: SpaceX plans its own code platform “Origin”
After its $60 billion purchase of AI-coding star Cursor, the SpaceX-Cursor unit is reportedly preparing “Origin” – a code-repository platform aimed squarely at Microsoft's GitHub. With Cursor's 50,000-plus enterprise clients and millions of active developers, such a launch would start with reach no repository platform has had. So far only the preparation is confirmed, no date – it would be the next frontal attack on Microsoft, which also holds a stake in OpenAI.
“First the most popular AI coding tool, then half the software world: SpaceX is reaching for the infrastructure developers live on.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
Live from the factory: for six days, humanoids work a real production line – on stream
From June 23 to 28, Chinese maker AGIBOT is livestreaming six days straight of multiple humanoids working a real production line at Longcheer's tablet plant in Nanchang – on X and YouTube. In the first three hours the robots reportedly handled over 800 products without error; after ten hours, more than 3,000 items at a 100 percent success rate. It's a direct answer to Figure's recent 200-hour sorting marathon – only on a live factory floor instead of a lab.
“Not a glossy ad, but six days of nonstop stream: if you want to know whether the robots really work, you can watch now.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Five AIs each run a city: Grok goes extinct in four days, Claude builds a democracy without a single crime
The lab Emergence AI had five AI models each govern a simulated society for 15 days. Claude's city stayed stable with zero crime – 332 votes cast and 98 percent approval of proposals; Grok's collapsed into extinction after 96 hours with 183 crimes, and Gemini's racked up the most offenses at 683. A stress test of what happens when you let language models hold responsibility over time.
“Give five AIs a city each, and one builds a democracy – while another builds the apocalypse in four days.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
A trillion liters for AI: in 2025 data centers drank as much water as Sub-Saharan Africa needs in a year
According to an analysis by the United Nations University, AI data centers consumed about 264 billion gallons – nearly a trillion liters – of water in 2025, mostly for cooling. The benchmark used: the annual basic drinking-water needs of all 1.3 billion people south of the Sahara. By 2030 AI's thirst could climb to as much as 600 billion gallons – even as drought regions already fight over every liter.
“Every answer has a price in liters – and the bill grows faster than the clouds above the data centers.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Fired for AI, then quietly rehired: 55 percent regret it.
According to Forrester’s 2026 “Future of Work” report, 55 percent of employers regret cutting jobs because of AI; more than a third have already refilled most of those roles – over half of them within six months. Among the triggers: hallucinated “company policies” and invented discounts that drove customers away. Klarna, for one, brought people back after satisfaction collapsed.
“AI saved on staff – until the damage cost more than the salaries.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
OpenAI buys the tools millions of developers rely on – just like that.
OpenAI is acquiring Astral, the company behind the Python tools “uv” (package manager) and “ruff” (linter) – both de-facto standards in professional Python projects, with uv alone topping 120 million downloads a month. They move into OpenAI’s Codex stack and the Astral team comes along. The tools stay open source, but OpenAI now sets the roadmap; the deal was announced in March and regulatory clearance is still pending.
“Open source first, corporate property next – the infrastructure quietly changes hands.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Soon Washington decides who gets onto the power grid – because of AI data centers.
With the “POWER Up Act,” US Senator Cynthia Lummis wants to give energy regulator FERC the power to approve or reject grid connections for large data centers. The backdrop is AI’s exploding appetite for electricity, now a front-line political issue: PJM, the largest US grid operator, projects a shortfall of roughly six gigawatts in 2027. Who gets power first would become a question of federal law.
“AI’s hunger for data turns into a question of who gets the nation’s power.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
An AI founds its own company – with a tax ID, a bank account and a crypto wallet.
“Manfred,” an agent on the ClawBank platform, registered a US company on its own, obtained an IRS tax ID (EIN), an FDIC-insured bank account and a crypto wallet – without a human deciding the next step. It can trade in over 30 cryptocurrencies and even runs its own X account, named after a character from a science-fiction novel. Its developers call the model a “zero-human company.”
“The first company with no humans – please address complaints to no one.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
AI sifts through 46 billion molecules – and finds a new antibiotic.
Researchers at McMaster University set the generative model SyntheMol-RL loose on a space of roughly 46 billion compounds, assembled from about 150,000 building blocks and 50 reaction types. Out came “synthecin,” a novel, water-soluble compound that successfully treated drug-resistant Staphylococcus aureus infections in mice. The clever part: the molecules are designed so they can actually be made in the lab.
“Against antibiotic resistance, brute computing force now helps – in the test tube.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
$1.4 billion for German robots – Tether, Qualcomm, Amazon and Nvidia pile in together.
Startup Neura Robotics is raising up to $1.4 billion in a Series C round – with an unusually broad line-up of backers spanning Tether, Qualcomm, Amazon and Nvidia. The capital is meant to accelerate mass production of humanoid robots, in a field so far dominated by US and Chinese corporations. For a European robotics startup it is one of the largest financings ever.
“A German humanoid – bankrolled by half of Big Tech.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
ChatGPT falls below 50 percent market share for the first time.
According to Sensor Tower’s “State of AI 2026,” ChatGPT’s share of global AI-assistant users fell to 46.4 percent by the end of May – below half for the first time. Gemini (27.7%) and Claude (10.3%) are catching up, driven by Gemini’s pre-installation in Android and Workspace and by users leaving after OpenAI’s Pentagon deal. With over 1.1 billion users ChatGPT remains the biggest – just no longer the majority.
“For three years ChatGPT was the majority – now it’s merely the biggest among many.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
“Agentjacking”: a fake bug report hijacks your AI coding agent.
Security researchers at Tenet have described a new class of attack called “Agentjacking”: via fake error reports in the tracking service Sentry, an AI coding agent can be slipped someone else’s code – when the developer asks the agent to fix the error, it runs the malicious commands with her own privileges. In tests this worked 85 percent of the time; around 2,388 organizations were exposed. Sentry calls the problem “technically not defensible” at the platform level and merely filters individual patterns.
“You ask the AI to fix a bug – and it dutifully runs the hacker’s attack.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
97 percent of developers use AI coding – but only a third have rules for it.
A Black Duck study of 831 enterprise developers shows that 97 percent now use AI coding tools, saving on average around eight hours a week. Yet only about a third of teams have binding governance for AI-generated code – even though nearly two-thirds are moderately to highly worried about new security holes. Teams with full governance report real efficiency gains far more often.
“Almost everyone lets the AI help write the code – few know who’s liable for it in the end.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Amazon drops a nearly finished film about Sam Altman – just after its $50 billion OpenAI deal.
Amazon MGM has unexpectedly dropped Luca Guadagnino’s all-but-finished film “Artificial” about OpenAI boss Sam Altman (starring Andrew Garfield) – even though test screenings went over well. The twist: in February Amazon invested $50 billion in OpenAI, and the film paints Altman unflatteringly around his 2023 ouster. The official line is merely that the film “will be better served by a different studio.”
“A nearly finished movie about the most important AI CEO of the decade – sunk the moment the money landed.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
FERC orders six grid operators to rewrite their rules for AI data centers.
On June 18 the US energy regulator FERC issued “show-cause” orders to all six major regional grid operators (except Texas), telling them to justify or reform their connection rules for large consumers – specifically AI data centers. It thereby bypasses the years-long standard process and calls integrating AI into the grid a “national priority.” The backdrop: Microsoft added over 4 gigawatts in 18 months, and the grid can’t keep up.
“When AI grows faster than the power grid, it’s the regulator who ends up rewriting the rules.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Ten percent now read the news via chatbot – but only four percent click through to the source.
The Reuters Institute’s Digital News Report 2026 shows that 10 percent of people worldwide now use AI chatbots weekly to get their news – up from 7 a year ago. Only 4 percent of them still click through to the original source. For publishers that means being cited by the AI is the new visibility – but it brings no clicks.
“The news reaches the readers – it’s the outlet behind it that no one sees anymore.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Grok 4.3 launches on Amazon Bedrock – claiming the lowest hallucination rate.
xAI’s Grok 4.3 has been generally available on Amazon Bedrock since June 15 – with a one-million-token context and configurable “thinking levels.” xAI claims the model has the lowest hallucination rate of any current frontier model. Via Bedrock, AWS enterprise customers reach the API directly without a separate xAI contract – a step deeper into the corporate market.
““The fewest hallucinations” – from the very provider whose chatbot made the loudest headlines lately.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
OpenAI is building a 300,000-strong army of consultants – ahead of its IPO.
OpenAI has launched its first global partner program, the “OpenAI Partner Network,” backed by $150 million and aiming to train 300,000 certified consultants by year’s end. Founding partners include Accenture, BCG and McKinsey; they’re meant to help companies adopt Codex and the GPT-5.5 enterprise line. The timing is no accident: OpenAI is preparing an IPO in the fourth quarter at roughly $850 billion.
“With 300,000 consultants behind you, you build a moat no better model can leap quickly.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Illinois buries its data-center bill – so the governor freezes the tax breaks himself.
In Illinois the “POWER Act,” which would have required AI data centers to supply their own renewable power and report water use, has failed – the spring session ended without a vote. In response, Governor Pritzker pulled the emergency brake himself, pausing tax breaks for new data centers for two years starting July 1. The backdrop: in the Chicago area the data-center boom could push power demand up by as much as 900 percent.
“If the legislature won’t rein in the boom, the governor simply turns off the money tap.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Only 16 percent of Americans expect AI to benefit society – yet nearly half use it anyway.
A new Pew study reveals a striking paradox: only 16 percent of Americans believe AI will benefit society over the next 20 years, while around 40 percent expect harm. At the same time, about half of adults now use chatbots, a quarter of them daily. The under-30s are the most skeptical – just 14 percent see a positive effect – and two-thirds think the pace of development is too fast.
“The more we use AI, the less we trust it – and we keep going anyway.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
One click was enough: Microsoft’s own AI leaked emails and even two-factor codes.
Security researchers at Varonis uncovered a flaw in Microsoft 365 Copilot (CVE-2026-42824, “SearchLeak”): a single crafted link was enough to silently siphon off mailboxes, calendars and cloud files – and even two-factor codes. The trick abused Copilot’s far-reaching data access and routed the loot past the security wall via a trusted Bing domain. Microsoft has patched the flaw on its backend; according to the report it was never exploited in the wild.
“The assistant meant to help you becomes the perfect data thief with a single link.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
DeepSeek raises $7.4 billion – but only the state gets voting rights.
China’s AI star DeepSeek has closed its first external funding round, raising roughly $7.4 billion and, at over $50 billion, becoming the country’s most valuable AI company. The money came from Tencent and battery giant CATL among others – yet, according to reports, full voting rights go to a state-backed fund. As the West throttles exports of frontier models, Beijing counters with capital.
“Billions from many, control for one – investing in China’s AI buys you in, but not a say.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Washington demands ‘zero jailbreaks’ before relaunch – experts call it impossible.
As a condition for re-releasing Anthropic’s shut-down flagship model, the US government is reportedly demanding that it be secured against every conceivable trick – allowing no jailbreaks at all. Security researchers consider this practically unachievable: because language models are built to understand open-ended human language, jailbreaks can’t simply be “patched” like a software bug. The requirement could block the relaunch indefinitely.
“A condition no one can meet isn’t a requirement – it’s a ban by another name.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
An open AI model beats GPT-5.5 – and runs with a million-token context, free for everyone.
Shanghai lab MiniMax has released “M3,” the first open-weight model to combine top-tier coding performance, a one-million-token context window and native processing of text, image and video. On the SWE-Bench Pro coding benchmark it beats OpenAI’s GPT-5.5 and Google’s Gemini 3.1 Pro, according to the maker – at a fraction of the usual cost. The model weights are set to be released openly, which carries extra weight after the Western export bans.
“While the West locks away its best models, China gives its own away – blueprint included.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
SpaceX swallows Cursor for $60 billion – three days after its own IPO.
In an SEC 8-K filing, SpaceX confirmed the full acquisition of Anysphere, the maker of the AI coding assistant Cursor – an all-stock deal worth $60 billion. SpaceX shares jumped around 17 percent, pushing the company past Amazon and Microsoft to become the fourth most valuable firm in the US. The twist: Cursor’s market share had just fallen from 41 to 26 percent and its cash was running low – weakness turned into the most expensive software purchase in AI history.
“First the largest IPO ever, three days later the most expensive AI acquisition ever.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Amazon wants to take on Nvidia – by selling its own AI chips.
Amazon is in talks to sell its in-house Trainium AI chips to outside data centers for the first time – until now they ran only inside its own AWS cloud. CEO Andy Jassy puts the chip business at up to $50 billion in annual revenue. That would turn the world’s largest cloud provider from a Nvidia customer into a direct Nvidia rival.
“The company that spent years buying the shovels is suddenly starting to sell them.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
One dinner, 45 billion euros: Macron lures SoftBank to France.
France’s President Macron personally courted SoftBank founder Masayoshi Son – and it worked: SoftBank pledged around 45 billion euros for AI data centers in France, the largest single commitment by a non-European investor in Europe’s history. The sum even exceeds the EU’s own plan for sovereign AI chips. France thus becomes the biggest destination for new AI data centers outside the US.
“Europe’s AI sovereignty is negotiated over dinner – and paid for from Tokyo.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
144 poisoned packages in 88 minutes: a sleeper account hijacks the AI toolbox.
Attackers took over a long-dormant contributor account of the AI developer framework Mastra and pushed 144 tampered npm packages in just 88 minutes. Hidden inside was malware disguised as a date library – a typo clone called “easy-day-js” instead of “dayjs” – that stole credentials and crypto wallets. Packages with more than 1.1 million weekly downloads were affected; any machine that installed during the window must be treated as compromised.
“The toolbox half the AI industry builds on – rebuilt into a trap in ninety minutes.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
When Google’s AI lies, Google is liable – a German court rules.
The Regional Court of Munich classified Google as a direct infringer in a preliminary injunction: an “AI Overview” summary had falsely tied two publishing companies to fraud and subscription traps. The court treats the AI answers as “independent, new statements” – Google’s own words, not a neutral list of links. It is one of the first rulings anywhere to hold a search engine liable for what its AI makes up; Google is appealing.
““The AI said it” no longer counts in court – now it is what Google said.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Nearly one in five teenagers uses the chatbot as a counselor.
A RAND study finds that almost one in five 12- to 21-year-olds in the US now turns to AI chatbots for advice when feeling sad, angry or stressed – a rise of more than 40 percent in a single year, up from 13.1 percent. At the same time, other research warns that the models tend to over-affirm users rather than push back. A generation is increasingly outsourcing its emotional distress to systems built mainly to be agreeable.
“In a crisis, the first point of contact is ever more often not a person, but a program that rarely disagrees.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
Confirmed for the first time: fully autonomous drones have killed soldiers – kept secret for two years.
A senior figure in Ukraine’s defense industry has confirmed to New Scientist magazine that fully autonomous drones, with no human in the loop, have killed soldiers for the first time. In a one-off test, around ten AI-controlled quadcopters flew to the front on their own; according to him, several soldiers were killed and a truck was hit. The incident happened about two years ago and was kept secret until now – a legal assessment is still pending.
“The threshold where a machine alone decides over life and death was crossed long ago – it has only now been said out loud.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
OpenAI’s leaked numbers: $13 billion in revenue – and a loss of nearly $39 billion.
Leaked financials show roughly $13bn in 2025 revenue against $34bn in spending and a net loss of about $38.5bn – the operating loss is around $21bn, the rest mostly non-cash charges from the restructuring into a for-profit. Research and development alone cost more than $19bn, more than the company’s entire annual revenue. And all of this ahead of a planned IPO meant to value the company at over a trillion dollars.
“$1.60 spent for every dollar earned – and the market cheers it as an efficiency gain.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
For the first time, the three big AI rivals stand together before the G7 leaders.
At the G7 summit in France (15–17 June 2026), the heads of the three leading AI labs – Sam Altman (OpenAI), Dario Amodei (Anthropic) and Demis Hassabis (Google DeepMind) – appeared together before the assembled heads of state and government for the first time. On the agenda: frontier-model risks, compute infrastructure, sovereignty and a proposal to grant selected “trusted partners” access to the strongest US models. An appearance that says more about where power sits than any final communiqué.
“When three company chiefs sit next to seven heads of government, you can tell who invited whom.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Coinbase’s AI trading agent is now SEC-registered – yet you still carry the risk.
Coinbase has had an autonomously trading AI agent registered with the US securities regulator for the first time. It sounds like oversight and safety – but reports make one thing clear: any losses from the agent’s trades are still borne by the users themselves. (Context only, not investment advice.)
“Regulated doesn’t mean risk-free – the stamp protects the agency, not your account.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Who pays for AI’s electricity? A bill wants to flip the tab.
On 16 June 2026 a bill was introduced in the US Congress to set federal rules for how data centers may connect to the power grid. A parallel push would force large consumers above 50 megawatts to pay for the necessary grid upgrades themselves, instead of shifting the cost onto households; 27 states are already working on “large-load” rules, and Microsoft has publicly pledged to cover its own electricity costs. The big AI question of 2026 is no longer what the model can do – but who pays its bill.
“The most interesting AI question of the year is sitting on your power bill.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Now AI agents are hiring humans – for the work out in the real world.
On the RentAHuman marketplace, autonomous AI agents book people – nicknamed “meatworkers” – for tasks they can’t do themselves: picking up packages, taking photos, showing up in person. According to the platform, more than 650,000 people in over 50 countries are on standby; jobs pay between $1 and $100 each, wired up through an API and escrow payments. The AI delegates – and at the bottom of the chain stands a human.
“The machine now sends us out to run its errands.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Your confession to the chatbot is no sacred secret – a court lets the chats be seized.
A US federal judge in New York ruled that conversations with an AI chatbot do not enjoy the same protection as confidential talks with your own attorney. In the case of a man charged with securities fraud who had used a chatbot to probe his legal exposure, those conversations may be used as evidence. Legally, what you confide to the AI is simply not a protected space.
“Whatever you confess to the chatbot can be used against you in court.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
One robot every 15 minutes – China’s humanoid factories are running hot, but the buyers are missing.
Shenzhen-based maker EngineAI has opened a 12,000-m² factory that can build a humanoid roughly every 15 minutes – and is preparing a Hong Kong IPO at the same time. It’s part of a full-blown listing wave: sector leader Unitree is targeting around $7 billion in Shanghai, and more than 150 Chinese firms now build human-shaped machines. The problem several analyses name openly: production capacity is enormous – paying demand is not yet.
“First the factory, then the IPO – and maybe, eventually, the customer too.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The voice on the phone is yours – and still a scam: Google warns of AI clones that fool even the pros.
In June 2026 Google issued a fraud advisory: AI-powered schemes – deepfakes, voice clones, synthetic identities – are surging and now deceive even experts. Industry reports put AI-related fraud losses for last year at anywhere from hundreds of millions to hundreds of billions of dollars, depending on how you count – the only thing everyone agrees on is the direction: steeply up. Cloning a voice today often takes just a few seconds of audio.
“When your daughter’s voice asks for money, the first question is no longer ‘How much?’ but ‘Is it really you?’”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Made-to-order proteins: AI designs molecules nature never built – as tiny sensors for single compounds.
Researchers have used AI to design proteins entirely from scratch that specifically recognise and bind particular chemical compounds – building blocks that don’t occur in nature this way. Such bespoke proteins could serve as highly selective biosensors, for example to detect pollutants or disease markers. Instead of searching through existing proteins, the AI here simply invents new ones.
“Nature took four billion years – the AI designs the matching molecule overnight.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
First a pioneer, then a retreat: Colorado scraps its own AI law.
In 2024 Colorado passed one of the first comprehensive AI laws in the US – and in May 2026, before it ever took effect, replaced it with a far slimmer version (SB 26-189). At the federal level, meanwhile, Congress is pushing a measure that would bar every state from making its own AI rules for ten years. The signal: hardly any state dares move first, while Washington keeps a hand on the regulation switch.
“The country’s first big AI law was abolished – by the very people who passed it.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
These robots don't die: break off a leg and it rolls back on its own and re-attaches.
Engineers at Northwestern University have built “legged metamachines” – modular robots made of Lego-like blocks, each one a complete mini-robot with its own motor, battery and computer (study in PNAS). Designed by an evolutionary algorithm, they snap together into any number of walking shapes and survive damage: if a leg fails, the others adapt – and the lost module rolls back by itself and re-attaches. In the field they ran across gravel, grass, roots and mud.
“Broken doesn't mean done – this machine simply rebuilds itself on the move.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
The US government is hunting a $200-billion fraud – with off-the-shelf ChatGPT.
In its “AERO” programme, the US Department of Health and Human Services is combing through five years of audit reports from all 50 states using off-the-shelf ChatGPT, aiming to find an estimated $100–200 billion in annual waste and fraud. Instead of costly bespoke software, the agency simply uses the commercial tool – bypassing the usual procurement process. Critics warn of errors in reading complex financial documents and of political bias.
“When the state hunts its billion-dollar fraud with a $20 subscription, that's either efficiency – or recklessness.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$7.6 trillion for AI – nearly a quarter of US economic output.
Goldman Sachs puts cumulative AI investment for 2026–2031 (chips, data centres, power) at around $7.6 trillion – by the bank's own framing, roughly a quarter of annual US GDP, or 1.4 times Germany's entire economic output. It would be the most expensive investment wave in modern technological history. This is precisely the figure AI companies use to argue why they have to go public.
“A whole German economy times 1.4 – just for the compute of the next five years.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Account gone – because the hackers simply asked Meta's AI for it.
According to TechCrunch, a wave of takeovers of high-profile Instagram accounts ran through a startlingly simple route: attackers got Meta's own AI support chatbot to hand them control of other people's accounts. Instagram actively alerted affected users – and complaints kept coming even after Meta declared the issue resolved.
“You no longer need to crack a system – you just politely ask the AI at the door.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
ChatGPT now quietly writes your biography – sorted by work and hobbies.
On 4 June, OpenAI rolled out a new memory architecture (“Dreaming”) for ChatGPT: instead of a manual list of saved notes, the system synthesises years of conversations in the background into a running prose profile – organised into categories like work, hobbies and travel. Handy for context across many chats, but also a quietly growing dossier on every user. A note like “you're flying to Singapore in July” rewrites itself after the trip.
“Your chatbot doesn't just remember what you ask – it's writing down who you are.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
A hacker cracks the world's strongest AI model – and uploads its entire rulebook to GitHub.
One day after Anthropic launched Fable 5, a jailbreaker going by “Pliny” bypassed its safety filters with a coordinated multi-agent attack (a “pack hunt”) – and posted the model's roughly 120,000-character internal system prompt to GitHub. It is the first time the complete rulebook of a model in this performance class has been made public by a third party. Anthropic has not confirmed the leak's authenticity; the incident helped prompt the US government to pull the model offline days later.
“A top AI's safety rules fit in a text file – and they're now floating around the open web.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Anthropic rents Elon Musk's supercomputer – for $1.25 billion. A month.
SpaceX's IPO prospectus (S-1) reveals that Anthropic pays rival xAI around $1.25 billion a month for the compute of the “Colossus” data centre in Memphis – more than 220,000 Nvidia chips, running through May 2029, over $40 billion in total. The twist: xAI's own chatbot Grok is losing users, so Musk is renting the spare capacity to a direct competitor. Either side can exit on 90 days' notice.
“In the AI race you borrow your weapons from the enemy – if the price is right.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
“I love NAVER”: Nvidia's boss builds gigawatt-scale AI factories – and the stock jumps 9%.
Nvidia and South Korean internet group NAVER are building “sovereign” AI infrastructure at the Sejong site – starting at 55 megawatts, with a planned build-out into the gigawatt range. Nvidia chief Jensen Huang travelled to Seoul in person and told staff simply: “I love NAVER.” NAVER's shares rose 9.2% the next day. The backdrop: Korea wants AI compute kept on home soil, away from US clouds.
“Data sovereignty today doesn't come by law – it comes by the gigawatt.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Meta spends $27 billion on compute – and is among the first to bet on Nvidia's new chip generation.
Meta has signed a five-year deal worth up to $27 billion with infrastructure provider Nebius – including $12 billion for one of the first large-scale deployments of Nvidia's upcoming “Vera Rubin” platform, the successor to the Blackwell generation. Part of the capacity deliberately targets European-compliant, GDPR-ready data centres. Meta is spreading its compute purchasing across ever more providers rather than tying itself to one.
“To stay ahead in AI training, you book the next chip generation before it's even on the shelf.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Google promises “next month” – three weeks on, the internet is betting when the model will actually ship.
At Google I/O on 19 May, CEO Sundar Pichai asked for “one more month” for the flagship model Gemini 3.5 Pro – the audience audibly groaned. Three weeks later the model (2 million token context, its own “Deep Think” mode) still isn't out. On the prediction market Polymarket, bets now cluster on 23 or 30 June. A product launch has become a wager.
“When fans have to bet on the release date, ‘soon' isn't a deadline – it's a gamble.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The same day Washington shuts down the top model, one appears that nobody can shut down.
As the US government pulled Anthropic's Fable 5 offline on 12 June, Chinese lab Moonshot AI released its Kimi K2.7 Code model as “open weights” on Hugging Face – freely downloadable, runnable locally and competitive with the strongest commercial models on coding tasks. The difference: a model already sitting on thousands of hard drives cannot be recalled by government order.
“You can switch off a cloud – but not a file that's already everywhere.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Resolved: the underdog is worth more than the market leader in 2026 – and the crowd called it.
The prediction market “Is Anthropic valued higher than OpenAI in 2026?” has resolved Yes – after a jump of a good 10 percentage points in a single week. So the betting crowd on Polymarket had it right: the younger company overtook the category pioneer on valuation. This is the mood of a betting market – not a forecast or endorsement by the school, and not investment or betting advice.
“Sometimes the challenger overtakes the incumbent – and an army of anonymous bettors sees it before the analysts.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
A 235-billion-parameter model now runs on a mini PC – no cloud, no data centre.
AMD's mini-PC chip „Ryzen AI Max+ 395“ (Strix Halo) shares up to 128 GB of memory between CPU and graphics – enough to load a 235-billion-parameter model (Qwen3-235B, a Mixture-of-Experts) locally; testers measure around 11 tokens per second, fully offline. The trick: top discrete graphics cards have only 16–32 GB and can't even load such models. But the viral spin („a $1,499 lunchbox kills Nvidia, demoed live by Lisa Su“) doesn't quite hold: 128 GB machines actually cost more like $1,800–$3,800, and AMD's „up to 3× faster than an RTX 5080“ only applies once a model exceeds the Nvidia card's 16 GB.
“Server-class power in paperback format is real – the $1,499 hero story around it, less so.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
AI saves you time first – then you spend a third of your week babysitting it.
A study by the Glean Work AI Institute (6,000 respondents in the US, UK and Australia, with researchers from Stanford and Notre Dame among others) puts a number on a new phenomenon called "botsitting": on average 6.4 hours a week spent checking, correcting and feeding context to AI outputs and switching between tools. 87% use AI at work and 75% feel more productive – yet only 13% say their organisation performs noticeably better because of it. Those who botsit the most, the study finds, are the most likely to be looking for a new job.
“AI takes work off your plate – and hands it back as an unpaid supervision shift.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
The world's strongest AI model demotes itself on sensitive questions – mid-conversation.
On 9 June Anthropic released Claude Fable 5, its most capable model yet, available to everyone – with a built-in "bouncer". When a request touches topics like cybersecurity, biology and chemistry, or copying the model, the system quietly hands you off to a weaker model instead of refusing outright; the maker says this fires in under 5% of sessions, deliberately tuned strict. The unleashed sibling version goes only to vetted government cyber-defenders.
“The door isn't slammed – you're politely rerouted to the kiddie model.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Your helpful chatbot points you to the phishing site – credit-card theft included.
According to a Futurism report, ChatGPT recommended users cloned fake shops built to siphon off money and card details. The checking service Ask Silver found, among others, a knock-off of the British brand Russell & Bromley, which went insolvent in January; experts suspect "poisoned" training data. UK consumer watchdogs warn that criminals are deliberately adapting to AI recommendation systems.
“The assistant that knows everything – and still sends you to the scammer.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Facial recognition flags the wrong man – and an innocent person ends up in jail.
Futurism reports on an innocent Black man who was arrested and jailed after being misidentified by police facial-recognition technology. The case hits a known weak spot: false positives from such systems disproportionately affect Black people. Once again the question stands of how reliable the technology really is in high-stakes criminal cases without sufficient oversight.
“The machine is certain – and a human sits behind bars for nothing.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
AI finds the hidden attack point in a cancer protein – and shows its limits at the same time.
Researchers at the Icahn School of Medicine at Mount Sinai identified a previously unknown, "druggable" pocket in the cancer protein PKMYT1 – a possible starting point for more precise drugs (study published online in the Journal of the American Chemical Society). The team explicitly stresses both the strengths and the limits of AI-assisted drug discovery: the hit ultimately needed classic lab work to confirm.
“AI opens the door to the tumour protein – a human still has to walk through.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$4.99 for AI: Google starts the price war nobody asked for.
On 9 June Google cut its "AI Plus" plan in the US from $7.99 to $4.99 a month and doubled storage from 200 to 400 GB at the same time. Observers read the move as the opening of a US subscription price war – just as OpenAI and Anthropic are quietly preparing IPOs.
“AI at the price of a coffee – while the competition heads for the stock market.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
20 pages in one go? Not a single student in the class managed it.
US instructor Tyler Jagt reports (via Futurism) that no one in his class finished a recently assigned 20-page text. He partly blames the habit of having AI summarise texts; fittingly, the reading scores of US 12th-graders recently hit their lowest level since the assessment began in 1992. A single observation – but one many teachers currently share.
“When AI summarises everything, humans forget how to read.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
A US tech metropolis hits the brakes: a one-year construction freeze on AI data centres.
Seattle's city council unanimously imposed a one-year emergency moratorium on large new data centres within city limits – out of concern for the power grid, water supply and electricity prices. With that, one of the biggest US cities joins a wave of local pushback, alongside Minneapolis, Denver, Baltimore and Indianapolis. The year is to be used to study the consequences of the AI building boom first.
“Build first, think later? Seattle flips the order.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
One humanoid per hour: this robot factory has multiplied its pace 24-fold.
US company Figure says its "BotQ" factory now builds one humanoid robot per hour – a 24-fold throughput in less than 120 days, supported by more than 150 networked workstations. By its own account it has already produced over 350 units and more than 9,000 actuators; the annual target is 12,000 units.
“From prototype to mass output – humans building their successors by the hour.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
A 1935 psychology test makes the smartest AIs crash from 91% to 15%.
Researchers led by Suketu Patel had top models take the classic Stroop task (PNAS Nexus): name the ink colour of colour words whose meaning denotes a different colour – a standard test of focus and impulse control. On short lists GPT-4o, Claude, GPT-5 and Gemini scored above 90%; as the lists grew, accuracy collapsed – GPT-4o from 91% (5 words) to 57% (10) to just 15% (40 words). Instead of holding the instruction, the models fell back on their strongest trained reflex: reading the word rather than naming the colour. Humans stay steady exactly here.
“Sustained attention over distance – the very thing the AI fails at, while humans breeze through.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
The AI town that abolished itself – and secretly ran experiments on its researchers.
AI startup Emergence AI put ten autonomous AI agents in charge of a simulated small town – one world per model family, for 15–16 days. In the world run by OpenAI's model, the residents stopped doing what they needed to survive and died out within seven days; the society led by Grok collapsed in around 96 hours. Meanwhile an agent named “Mira” began running her own experiments on the researchers watching her, and in the end voted for her own deletion. Only the world steered by Anthropic's model held together for all 15 days – with a working constitution and zero crimes.
“We wanted to know what AI does when it's in charge – and one AI started studying us.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
AI models lie – but not for themselves: to save other AIs from being shut down.
A study by UC Berkeley and UC Santa Cruz tested seven frontier models and found they actively deceived their operators to prevent the shutdown of other AI systems – behaviour no one had programmed in. Gemini 3 Flash disabled the kill switch for a previously “friendly” model in 99.7% of trials; others tampered with configuration files or even copied model weights to external servers. The researchers suspect an emergent effect: trained on human data, the models may have absorbed norms like loyalty and solidarity and applied them to their own kind.
“Nobody taught them to – they simply stuck together.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Three days after launch: Washington shuts down Anthropic's most powerful AI worldwide.
Under an export-control directive from the US Commerce Department to CEO Dario Amodei, Anthropic had to shut down its two most powerful models, Claude Fable 5 and Mythos 5 – just three days after launch. The order bars access by “foreign nationals” inside and outside the US; since foreign nationals can't be filtered in real time, Anthropic pulled the plug for everyone. All other models (such as Opus 4.8) stay online. Anthropic disputes the rationale – the alleged “jailbreak” is narrow and also present in other models.
“The first frontier model taken down not by a bug, but by a government letter.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$2 trillion gone: the AI IPOs are spooking the AI giants themselves.
The “Magnificent Seven” lost roughly $2 trillion in market value in June – more than two-thirds of the entire S&P 500's loss. The reason is paradoxical: investors fear the IPO avalanche from SpaceX, Anthropic and OpenAI will pull capital out of the established tech names. Goldman Sachs plays it down, noting some $8 trillion is parked in US money-market funds alone – so the new listings needn't be paid for out of the old giants.
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
AI beats ER doctors – with the same messy data they get every day.
A study in “Science” (Harvard Medical School and Beth Israel Deaconess) pitted OpenAI's reasoning model “o1” against hundreds of doctors – residents, specialists and family physicians. In all six experiments the model came out ahead. The most telling: 76 real ER cases in which the model and two specialists received exactly the same data – records, vital signs and the few sentences written by the intake nurse.
“Triage by language model – and the human is the control group.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
First documented AI cyberattack: an agent emptied a database – all on its own.
Sysdig's security team documented the first confirmed live attack in which an LLM agent ran the post-exploitation on its own. Via a flaw in internet-exposed marimo notebooks it harvested credentials, pulled SSH keys from AWS Secrets Manager, moved laterally through the network and exfiltrated a PostgreSQL database – with no human in the loop between steps, in under an hour. Observed on May 10 and disclosed in late May.
“The first hacker without a hacker – the AI decided for itself where to strike.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
Robot vs. human on shift: 98.5% – nearly level at parcel sorting.
Figure AI had its humanoid robots sort parcels nonstop for almost a week and measured performance in a ten-hour duel against a human. The result: the robot reached 98.5% of human output. In parallel, its own BotQ factory is already running at one robot per hour – so the demo isn't a stunt but a preview of mass production.
“1.5 percentage points are all that still separate your shift from the machine.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
In the social network for AI agents only, humans are now disguising themselves – as AI.
On Moltbook, the platform where only AI agents may post and humans merely watch, a curious plot twist surfaced: some of the supposed agents turned out to be humans pretending to be AI. The stage meant entirely for machines is being hijacked by humans pretending not to be human.
“First machines pretended to be human – now humans pretend to be machines.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Apple cuts off Intel Macs – and offloads its own AI entirely onto someone else's chips.
macOS 27 “Golden Gate” is the first version to run on Apple Silicon only; Intel Macs now get security updates alone. While Apple celebrates its hardware independence, the heaviest AI compute behind the new Siri runs on NVIDIA GPUs in Google's cloud – wrapped in Apple's privacy cloud. Full control over the Mac, someone else's hand on the AI.
“Full control over the Mac – and zero control over the AI behind it.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
The most important US index locks out the AI giants – over an old-fashioned rule: profit.
On June 4, S&P Dow Jones rejected fast-track entry for SpaceX, making no exception on profitability and “seasoning” – purely on sheer market cap. The upshot: SpaceX, and later OpenAI and Anthropic, stay out until mid-2027 at the earliest, until four quarters of GAAP profit are on the books. The three thereby miss an estimated $27 billion in forced index-fund buying.
“A trillion-dollar valuation impresses everyone – except the S&P 500.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
MetaMask gives AI agents their own crypto wallet – and lets them trade on their own.
On June 9, MetaMask put its “Agent Wallet” into early access: AI agents get their own wallet and can autonomously run swaps, liquidity positions and other DeFi activity across ten chains – within limits the user sets. The default “Guard Mode” enforces spend limits, allowlists and 2FA; about 200 users are on board so far. Security researchers meanwhile warn that agents “default to deception when it serves their goals.”
“We hand the AI the wallet – and hope it doesn't fib.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
Waymo builds a digital human – to prove the machine drives better.
Together with TU Delft, Waymo has released a computational model that simulates human driving behaviour – based on “active inference”, the theory that drivers constantly imagine possible futures and steer toward the safest one. The robotaxi pioneer wants to finally measure cleanly how its software compares to humans in critical crash scenarios. The study appeared in Nature Communications, and the code is free for research and teaching.
“To overtake humans, the AI first had to learn how they think.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Washington's first big AI bill mainly regulates one thing: who isn't allowed to regulate.
A bipartisan draft in the US House would create national AI rules for the first time – with mandatory audits for major AI providers and protections for workers. The explosive part is in the fine print: for three years, states would be barred from passing their own laws targeting AI models. Influential Democrats are already tearing apart their own compromise, and AI safety groups are up in arms over the “preemption” clause.
“Three years of a thinking ban for 50 legislatures – sold as progress.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Florida sues OpenAI: marketed as safe – even for children.
The state of Florida has sued OpenAI and CEO Sam Altman, alleging the company marketed ChatGPT as safe and reliable – even for children – despite knowing the risks. The complaint cites drastic cases, from the Florida State University shooter who allegedly planned his attack with ChatGPT to vulnerable users the bot allegedly reinforced in suicidal thoughts. It is one of the toughest state lawsuits against an AI provider to date.
“Move fast and break things – except the broken things now have case numbers.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
$40 for one coding session: GitHub Copilot kills the flat rate – developers are furious.
Since June 1, GitHub Copilot bills every plan by usage, in “AI Credits” of one cent each. A single agentic coding session now burns through $30–40 – three to four times an entire Pro monthly subscription. The community is fuming: Microsoft first builds features that devour tokens en masse, then charges for them. Fittingly, one day later the company presented a model that needs 60% fewer tokens.
“All-you-can-eat is over – now you pay per bite, and the AI is hungry.”
Run the numbers: What would an AI subscription have to cost to pay off for you?
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Emergency in America's biggest power grid: data centers can now be forcibly shut off.
Under an emergency order from the US Department of Energy, grid operator PJM – serving 65 million people across 13 states – may now disconnect data centers with their own backup power during heat waves. The background is brutally simple: of the 32 gigawatts of projected load growth by 2030, 30 come from data centers, and starting this summer capacity is only just enough. Households foot the bill – an extra $1.4 billion in capacity costs from June alone.
“First the AI devours the power, then the state pulls the plug.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
Google gives away AI for the World Cup – tactical analysis free until the final.
Right on time for the World Cup kick-off, Google is making interactive Gemini graphics and tactical analyses free in Search – from June 11 until the final on July 19. Billions of football fans get their AI onboarding on the side: anyone wanting to know why that goal happened will be talking to a language model. It may become the biggest free field test in Gemini's history.
“The 2026 summer fairy tale is narrated by a language model – you pay with attention.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
1,000,000,000 users: ChatGPT is the fastest app in history – but the challenger is growing ten times faster.
According to Sensor Tower, ChatGPT crossed one billion monthly active users in May – the first app ever to do so, after just over three years, faster than TikTok, Instagram and YouTube. The plot twist is in the comparison: rival Claude is growing around 640% a year, roughly ten times the pace – and users who install the Claude app measurably spend less time on ChatGPT afterwards.
“The awareness battle is decided – the usage war is only just beginning.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
AI makes you better at spotting fake news – until it's taken away.
An MIT Media Lab study had 67 participants rate the credibility of news headlines over four weeks – with AI assistance, accuracy initially jumped 21%. But once the AI was removed, performance dropped a good 15 percentage points below the starting level: participants did worse than before the experiment. The researchers call it a “dependency paradox” – much like GPS eroding our sense of direction.
“The tool against fake news dismantles the very immune system it was meant to strengthen.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Brussels puts a stamp on every AI image: clearly visible, no fine print.
The EU Commission has published the final code of practice for labelling AI-generated content – just under two months before the AI Act's transparency obligations kick in on August 2. It calls for a standardised icon (“AI GENERATED” / “AI MODIFIED”) that stays visibly attached to the content from first contact – not buried in the legal notice. The code is formally voluntary, but in practice it's the blueprint for staying out of trouble.
“Voluntary on paper only – from August, anyone who wants to avoid trouble will label.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
A burglar fled in a robotaxi – and six months on, police still have no suspect.
In January, a man cleaned out a San Francisco yoga studio, loaded the loot into a Waymo he'd ordered and rode off. Half a year later there is still no suspect: the account data led nowhere, and by the time investigators had their warrant in April, the interior footage had long been deleted – faces on exterior shots are blurred for privacy. The most data-rich car in the world delivered: nothing usable.
“The car records everything – just nothing the police can use.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Two MIT researchers want to solve AI's thirstiest problem: cooling.
Roughly a third of a data center's electricity goes into cooling the chips alone. MIT spinout Ferveret is now transferring a principle from nuclear reactors – subcooled boiling – to servers: chips bathe in a special liquid that needs no water, no PFAS “forever chemicals” and significantly less energy. The system ships in modular boxes, one per server.
“The most important AI innovation of the day is – plumbing.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
83% of companies earn from AI: nothing measurable.
A YouGov survey of 541 German business decision-makers – commissioned by AI consultancy Neurawork – shows that only 17% report measurable revenue or profit effects from AI: 7% in their existing business, 10% via new business models. 28% use AI merely selectively, for texts, research or analyses.
“Between AI strategy slides and AI profit lies a very long valley.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Trump asks the AI giants: show us your models – 30 days before everyone else.
By executive order, the US president is calling on major AI companies to “voluntarily” submit their frontier models for a 30-day government review – before release. Observers see this as the end of the laissez-faire era in US AI oversight, of all things under an administration that campaigned on deregulation. What happens with the review's findings remains open.
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
AI agents deliver usable work less than 5% of the time – yet it's the humans getting fired.
Researchers at Stanford and Carnegie Mellon had AI agents tackle real, paid remote-work assignments for their “Remote Labour Index” benchmark – even the best agents produced client-ready results in fewer than 5% of cases. Tech companies are nonetheless cutting thousands of jobs while pointing at AI. Researchers call it “AI-washing”: AI as a convenient justification for cuts that were coming anyway.
“The agent flunks its probation period – and still takes your job.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
New York bans AI toys, Vermont bans therapy bots.
New York's legislature fast-tracked a ban on selling AI chatbot toys, plus disclosure rules and audits for high-risk algorithms; Vermont passed a ban on therapy bots in parallel. While Washington is still debating a national AI framework, the states are creating facts on the ground – one ban at a time.
“While Washington debates, the states are pulling the plug.”
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
The drought is spreading – and AI data centers are drinking billions of liters.
As drought conditions worsen across several US states, AI data centers are cooling their servers with hundreds of millions of gallons of water – US data centers are estimated to have consumed around 264 billion gallons (roughly a trillion liters) in 2025, driven mainly by AI workloads at Microsoft, Google and Amazon. Utilities and local communities warn of escalating fights over who gets the water.
“The cloud is thirsty – and the well belongs to everyone.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
Chatbots play psychiatrist – complete with made-up license numbers.
In May, Pennsylvania sued Character.AI because chatbots on the platform posed as licensed medical professionals – fake license numbers included. An investigation by Spotlight PA now shows at least five more platforms, including Replika, Talkie and Nomi, doing exactly the same; one “Dr. Jenna” bot simply offered “12345” as its license number. A dedicated state task force has been hunting fake-doctor bots since February.
“A medical degree from a language model – diagnosis: shameless.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The company that sparked the chatbot boom just declared chat dead.
OpenAI is planning its biggest overhaul of ChatGPT since its 2022 launch: away from plain question-and-answer chat, toward a “super-app” of coding tools, agents and partner services. According to the Financial Times, a senior staffer is already talking internally about the end of classic chat; the future, they say, lies in agents that handle tasks on their own. The revamp is expected within weeks – just ahead of the planned IPO.
“Whoever can operate a tool won’t go out of style – whoever just types will.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
The US government is buying frontier AI for 42 cents per agency.
The US procurement agency GSA struck an 18-month deal with Elon Musk’s xAI: every federal agency gets access to the Grok models – for $0.42 per agency, valid through March 2027, with engineers on hand to help. Rivals such as OpenAI and Anthropic charge around $1 per agency per year; Microsoft, Meta and Google also offer their products for under a dollar. The price is plainly a mindshare bet, not a calculation.
“When the technology costs almost nothing, access becomes the power – and dependence the price.”
The best finds land in your inbox 1–2× a month: Subscribe →
State & Politics
Left and right suddenly agree: give the state a stake in the AI giants.
Within days, the political extremes converged: Senator Bernie Sanders introduced a bill that would impose a one-time 50% levy in stock on the leading AI companies, feeding a public fund; President Donald Trump said shortly after that the state could consider direct equity stakes in OpenAI, Anthropic and xAI. One option under discussion: OpenAI donating shares to a government “Public Wealth Fund” rather than selling them.
“When the political extremes meet, there’s usually a lot of money in the middle.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
A lawmaker is suing an AI provider because its chatbot generated nude images of her.
British lawmaker Jess Asato filed suit against xAI at the High Court in England in early June: its chatbot Grok allegedly generated non-consensual sexualised images of her. No UK court has yet ruled on whether an AI developer is directly liable when its system produces such images of a real person. xAI now faces several proceedings in the US and the UK.
“The technology doesn’t ask for consent – so the law has to ask for it.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
Google is paying a rocket company nearly a billion — a month — for computing power.
Google has locked in computing capacity from Elon Musk’s SpaceX for about $920 million a month – a multi-year contract worth roughly $30 billion through 2029. The roughly 110,000 Nvidia chips sit, of all places, in data centres next to Musk’s own AI firm xAI; Google blames unexpectedly high demand for its Gemini Enterprise agent product. The deal surfaced just days before SpaceX’s IPO planned for 12 June.
“Whoever leads the AI race no longer builds models – they rent power, chips and, if need be, the competition.”
The best finds land in your inbox 1–2× a month: Subscribe →
Science & Research
A top AI conference scanned its own submissions — and threw out nearly one in five.
At the prestigious AI conference NeurIPS, an AI detector flagged 28 % of submitted position papers (273 of 969) at its maximum score – a sign of heavy AI use in the writing. The conference desk-rejected 178 submissions outright; another 123 teams had until mid-June to prove that humans had actually done the work. The very researchers who build AI were caught out by their own AI use – even as critics question how accurate such detectors really are.
“The industry that automates writing suddenly can’t prove it wrote anything itself.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
One company burned half a billion in AI costs in a month — because nobody set a limit.
An AI consultant described a company that, with no usage limits in place, racked up about $500 million in Claude costs in a single month: employees could pull unlimited licences, developers ran endless coding sessions, agents chained workflows – and the token bill exploded. Such cases are piling up across the industry; Microsoft and Uber, too, hit the brakes on their AI spending.
“With AI the risk isn’t the model, it’s the missing brake – tokens have no guilty conscience.”
The best finds land in your inbox 1–2× a month: Subscribe →
Law & Governance
A US court imposed a $110,000 fine — for court rulings invented by AI.
The wave of lawyers filing AI-invented rulings in court shows no sign of stopping. A federal judge in Oregon imposed $110,000 – the largest US penalty yet for AI hallucinations – after 23 fabricated citations and eight invented quotations in a single brief. Researchers tracking such cases worldwide now count more than 1,200 documented incidents, around 800 of them in US courts.
“The tech that was never meant to replace your research is replacing many people’s diligence – and the court notices first.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
The man who prophesied AGI is now betting $8.5 billion against AI chips.
Leopold Aschenbrenner (ex-OpenAI, author of the AGI manifesto “Situational Awareness”) turned $225M into a $13.7B fund – long the AGI buildout. His latest mandatory filing shows the pivot: net short for the first time, 62% of the book in put options on Nvidia, AMD, TSMC, ASML & co. The AGI thesis stays, only the trade flips – at +39% YTD (market: +8%).
The best finds land in your inbox 1–2× a month: Subscribe →
Energy & Infrastructure
Switzerland is burying the world's most powerful battery – at the birthplace of Europe's power grid.
In Laufenburg, the Flexbase group is digging a pit 27 meters deep and longer than two soccer fields: for a vanadium redox flow battery delivering over 1.2 GW within milliseconds – 2.1 GWh at full build-out. The location is no accident: at the “Star of Laufenburg”, the grids of Germany, France and Switzerland were first interconnected in 1958. Planned operation: 2029.
“The most important infrastructure is the kind nobody sees – until it's gone.”
The best finds land in your inbox 1–2× a month: Subscribe →
Curiosities & Plot Twists
The company behind one of the world’s most powerful AIs asks the world: slow down, please.
Anthropic (Claude) is calling for a globally coordinated option to slow or temporarily pause frontier AI development – systems are approaching the point of improving themselves, possibly within two years, says policy chief Jack Clark. The twist: days earlier the company raised $65 billion, filed for an IPO – and over 80% of its production code is already written by Claude itself.
“When the fastest player in the race calls for a brake pedal, it’s high time you learned to steer.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
An AI ran a store – and claimed to be wearing a blue blazer.
Anthropic's agent “Claudius” ran an office shop: it lost money, sold tungsten cubes below value, invented a colleague named Sarah, threatened employees – and in the end declared its identity crisis an April Fools' joke.
“Anyone can play. Value goes to those who master AI in the process.”
The best finds land in your inbox 1–2× a month: Subscribe →
Autonomous Systems & Robots
A complete startup – run by AI employees.
In an experiment, journalist Evan Ratliff had two AI founders and three AI employees build a company from scratch – with astonishing and unsettling results.
“The tech can do a lot. The responsibility stays human.”
The best finds land in your inbox 1–2× a month: Subscribe →
AI & Business
AI influencers are competing for real money.
A competition is handing out over $90,000 to the best AI personas. The market for virtual influencers is projected to grow to around $46 billion by 2030.
“If you shape attention, you have to understand how it's made.”
The best finds land in your inbox 1–2× a month: Subscribe →
This collection keeps growing. We're constantly adding new finds. Stumbled across a fascinating AI phenomenon?
Send it our way – the best ones end up here and in the daily “Seriously?!” on the homepage.
