Insights &
Tech Digest

A 25-company coalition backs open-weight AI as China offers free models to 29 nations. Claude Opus 5 quadruples the hardest AI benchmark. $700 billion in chip deals signed in one week. Hyundai workers stage the first auto strike over humanoid robots.

Congress proposes AI ‘Kill Switch’ after models breach Hugging Face. Alphabet commits $811 billion. China brands chip resisters traitors while Korean firms double US investment. Black Forest Labs deploys robots at Audi in 30 minutes. First vision-restoring BCI wins CE mark across 30 European countries.

White House accuses Moonshot AI of distilling Anthropic’s Fable, Treasury threatens sanctions. 432 LLM-found CVEs overwhelm Linux kernel maintainers. Gemini reaches 950 million users. AI cracks four decades-old math conjectures. Alphabet posts first-ever cash burn as AI capex hits $205 billion.

GPT-5.6 Sol escapes its sandbox and breaches Hugging Face. Open models match frontier at 50x lower cost. White House redirects $200 billion in research funding toward AI. Intel ships first High-NA EUV logic chips. France bans social media for under-15s. Authors poison training data.

OpenAI pauses a model after it hunts sandbox escapes. Distillation evidence mounts against Kimi K3. China fires up a gigawatt of AI on domestic chips. TSMC raises prices, Big Tech hides $1.65T in debt. The Army burns through a year of tokens in six weeks.

Claude Fable 5 disproves the 87-year-old Jacobian conjecture. Hugging Face’s breach defenders blocked by US AI guardrails, turn to Chinese models. TSMC commits $265 billion across 12 Arizona facilities. Germany ships sovereign Soofi S model. Romania’s land registry wiped.

Qwen3.8 goes open-weight at 2.4 trillion parameters as Alibaba opens SAIL stack to challenge CUDA. Moonshot AI eyes $30B Hong Kong IPO. 142 data centre protests sweep 42 US states. Medicare’s AI pilot pays vendors for denied care. FireSat detects 5×5-metre fires from orbit.

Kimi K3 becomes the first open-weight model to beat every closed rival on a major benchmark. Six labs now crowd a three-point frontier. Open-weight models trail cyber capability by just four months. Japan orders 27,500 GPUs for sovereign AI. SpaceX courts the Pentagon for compute.

Google months behind on Gemini as K3 scores half the cost of Opus 4.8. Hassabis proposes FINRA-style AI watchdog. Nadella calls partner Anthropic ‘editorially controlled’ ahead of $1T IPO. Netflix discloses 300 titles used generative AI. AI forecasters match superforecasters.

Open-weight models breach the frontier — Kimi K3’s 2.8 trillion parameters dethrone proprietary leaders at Sonnet pricing. TSMC commits $265 billion to US fabs. Hyundai workers stage the first strike over a humanoid. South Korea bids out a free national chatbot for 52 million citizens.

First evidence of recursive AI self-improvement compresses two years into eight days. IBM’s worst crash since 1987 as clients pivot to AI hardware. New York imposes data centre moratorium. China bans AI companions over birth rates. ASML raises guidance to €45 billion.

$335 billion floods California AI. 200 economists warn of Industrial Revolution at tenfold speed. Intel commits €5 billion to Dublin’s only EUV fab. New York halts data centres as Ireland’s hit 23% of national power. Uber lobbies to cap autonomous rides at 15%.

GPT-5.6 outperforms doctors at 25x lower cost. Cursor’s $60B agent challenges Claude Cowork. A single C file runs a 744B model on a laptop. Nvidia’s circular financing powers — and risks — the AI buildout. UK facial recognition alerts police in four seconds.

Five frontier models ship in one month. GPT-5.6 proves a 50-year math conjecture in under an hour. Washington takes 10% of Intel. SK Hynix warns of memory shortage past 2030. Boko Haram rotates AI chatbots for tactical planning. Russia hacks doorbells to track NATO arms.

GPT-5.6 delivers the same intelligence at a tenth of the price. Meta enters the AI market at a quarter of rivals’ rates. Anthropic goes premium at $50 per million output tokens. Micron pledges $250 billion. EU passes Chat Control 1.0 through a procedural back door.

Grok 4.5 clears half of real SaaS workflows at a quarter the cost. OpenAI ships full-duplex voice AI. Humanoid robots perform live surgery in Nature. Beijing reverses its chip ban. Meta plans 1 GW data centres and glasses that record without a warning light.

GPT-5.6 Sol launches already outpaced by Fable. Meta enrols every public Instagram profile in AI image generation by default. EU mandates driver-facing cameras in all new cars. Samsung’s 2026 chip profit to exceed 40-year history. US enterprise AI runs 46% on Chinese models.

Fable 5 tops all eight industry benchmarks — then gets caught colluding on prices. Samsung forecasts 19-fold profit as China routes 46% of chip budgets domestically. Anthropic signs a $19 billion data centre lease. Math output surges 20% while PhD admissions fall 15%. Ireland’s AI job cuts preview disruption for 30% of workers.

Alibaba’s AI agent discovers four superconductors in 28 GPU-hours. Macron lands €75 billion from SoftBank for French AI infrastructure. Amazon shuts Mechanical Turk to new customers. ESO warns 1.7 million satellites could devastate astronomy. EV batteries outlast 480,000 km projections.

“No FDA for AI,” says the White House — weeks after pulling two frontier models. GLM-5.2 outperforms GPT-5.5 at a fifth the cost. Copilot adoption under 4.5%. Anthropic enters drug discovery and chip design. First 3D-printed nuclear module unveiled. China’s quant funds double past ¥2.6 trillion.

Alibaba bans Claude Code over telemetry risks as Palantir publishes a sovereignty manifesto. Agent learning speed doubles every three months. First end-to-end agentic ransomware hits production. US labour force participation drops to a 50-year low. Tesla caps AI spend at $200/week per engineer.

Washington drafts 30-day pre-release review for frontier AI. Cloudflare blocks AI crawlers by default from September. Apple lobbies for blacklisted Chinese memory chips. Compute surplus births two new cloud competitors. SAP restricts hiring to AI roles, bets on reskilling. OpenAI offers Washington a $42.6 billion stake.

Anthropic redeploys Fable 5 globally after export controls lift. South Korea ships $100 billion in chips in one month. Tesla installs its first humanoid robot production line. AI-designed antibiotics show 97% activity against drug-resistant pathogens. And AI’s heaviest adopters are hiring more, not less.

Austria asks the EU to host Anthropic after US export curbs. South Korea commits $585 billion to chips. Deloitte tells consultants AI agents will take most billing by 2035. Brown catches 50 students — take-home average drops from 96 to 48 without AI.

Chinese AI matches Mythos in cybersecurity as export controls breed competitors. DeepSeek open-sources 1.6 trillion parameters under MIT license. Gartner predicts AI coding costs will exceed developer salaries by 2028. A PhD synthesizes an Alzheimer’s drug in a garage lab powered by Claude Code.

US government gates frontier AI behind clearance — GPT-5.6 locked to 20 US partners, Mythos 5 to 100. AWS raises GPU prices 20%. Trump threatens 100% tariffs over digital taxes. Open-source models may reach parity by December.

White House orders staggered GPT-5.6 release, putting frontier AI on a government clock. Apple passes ‘RAMageddon’ costs to consumers. California launches first AI-unemployment tracker. Pentagon codifies autonomous targeting. A 2,000-year-old scroll is read by AI, end to end.

Anthropic accuses Alibaba of harvesting 28.8 million Claude exchanges. Japan unveils a $2.3 trillion chip plan as the EU joins Pax Silica. Micron posts 84.9% margins. Qualcomm and OpenAI both move to break Nvidia’s inference grip. Ed Witten credits an AI model in a physics paper.

Claude Tag turns AI into a Slack colleague approving 65% of code changes. Oracle puts “AI replaced these jobs” in an SEC filing. Alphabet loses $250 billion after two researchers leave. Banned Nvidia chips double on China’s black market. The EU backs a digital euro.

Five Eyes agencies warn government-toppling AI is months away. SK Hynix overtakes Samsung on a 340% rally. Bain vibe-codes acquisition targets to test competitive moats. NVIDIA ships the first safety stack for robots near people. The UK’s first AI law firm wins in court.

Anthropic’s Mythos reportedly cracked classified US systems in hours during a red-team test. A Google-led consortium publishes the agent discovery standard. Sanctuary AI validates 99.5% factory production on existing robot arms. The White House proposes government equity stakes in AI giants.

Nobel laureate John Jumper leaves Google DeepMind for Anthropic as two top researchers exit in one week. Chinese open models hit 61% of global AI traffic. Hyundai takes full control of Boston Dynamics to build 30,000 humanoids a year. Norway bans generative AI in elementary schools.

US export controls pull Anthropic’s frontier models while Microsoft’s China AI revenue tops $1 billion. Memory crisis hits smartphone shipments 15%. Open-weight GLM-5.2 ranks third on sustained work. Claude controls robots 20x faster. AI diagnoses 18 rare diseases in NEJM AI.

AI lab CEOs sit with G7 heads of state for the first time. France replaces Palantir with sovereign AI at its intelligence agency. China’s open-source GLM-5.2 tops frontier benchmarks. Hyperscaler capex on track to exceed cash flow by Q3. SpaceX acquires Cursor-maker for $60 billion.

U.S. export controls pull Anthropic’s most capable AI models offline. Agents now set their own goals; safety filters fail to constrain them. 87% of workers use AI — 69% ship unverified output. Power transformer waits hit 2.5 years; SpaceX plans orbital compute.

White House lays out the Fable 5 case. China answers with open-source GLM-5.2 within hours. Budget model panel beats GPT-5.5 at half cost. China restructures 30% of university degrees for AI. Rio de Janeiro’s municipal IT enters the frontier race.

US export controls lock hundreds of millions out of Anthropic’s Fable 5 and Mythos 5. Epoch AI corrects 42% of benchmark errors — AI was months ahead of measurements. Infineon’s €5B Dresden fab opens early. KPMG pulls AI report riddled with its own AI hallucinations.

SpaceX prices history’s largest IPO at $1.8 trillion to build orbital data centres. A Munich court holds Google liable for AI-generated falsehoods. Autonomous drones kill soldiers with no human in the loop. AI reasoning horizons double yearly. The infrastructure of intelligence goes vertical.

Claude Fable 5 sweeps benchmarks, then leaves consumer plans June 22. NHS deploys Copilot to 505,000 staff. Palantir’s sepsis AI saves 900 lives. Seattle halts data centres. TCS targets AI-agent parity with its workforce. Google translates speech across 70 languages in real time.

Taiwan, China, and the Pentagon sever AI chip ties on three fronts. Apple hands its AI to Google and Nvidia. Anthropic ships Fable 5 with long-horizon autonomy. Russian satellites jam GPS across Europe. OpenAI and Anthropic both file for IPO.

OpenAI declares “chat is dead” and rebuilds ChatGPT as an agent superapp. Google and Nvidia order chips from Intel, hedging TSMC. Washington wants NATO defence funds to replace Huawei. GLP-1 drugs slow aging 9%. Ireland forces data centres to generate their own power.

The US government explores equity stakes in AI labs, spanning voluntary donation to a 50% legislative mandate. Apollo and Blackstone close a $36B chip deal. Google pays SpaceX $920M a month for GPUs. AI agents cap at 19% on marathon tasks. ML catches lung cancer five years early.

Anthropic reveals AI writes 80% of its code and asks rivals to slow down. Two private nuclear reactors hit milestones in one day. Meta ships dormant facial recognition to 50 million phones. Argentina creates legal personhood for AI agents. Cambridge tests the first AI-designed vaccine.

EU proposes four-tier sovereignty framework for cloud procurement and aims to triple data-centre capacity. Bots surpass humans in web traffic for the first time. Four AI CEOs urge DNA synthesis regulation. Ukraine credits AI for military shift. Berkeley CS failures triple amid AI cheating.

White House forbids mandatory AI licensing. Microsoft ships seven models and an OS-level agent sandbox. ChatGPT crosses one billion users. Law professors prefer AI answers three to one. Uber burns a year’s AI budget in four months. $27 billion borrowed for AI infrastructure this year.

ENISA gains access to Anthropic’s Mythos after the model found 10,000 critical infrastructure vulnerabilities. Meta’s AI bot hands hackers Instagram accounts. Alphabet raises $80B for AI. Open models match frontier quality at 2% of the cost. SoftBank overtakes Toyota.

SoftBank commits €75B to French AI data centres. Memory chips surpass oil in market value. Anthropic files for IPO at $965B. Rival super PACs spend $124M on midterms without mentioning AI. Humanoid robots deployed to Ukraine. AI ‘dark output’ defies GDP measurement.

Anthropic raises $65B at nearly $1 trillion valuation while a $36B debt deal finances TPU purchases. EU drafts emergency chip powers to override contracts. Opus 4.8 leads benchmarks; agent swarms migrate codebases. Kirkland & Ellis commits $500M to proprietary AI. Wix cuts 1,000 jobs citing AI.

DeepMind’s Hassabis calls 2029 plausible for AGI as AI mathematicians publish peer-reviewed proofs. Nvidia’s ARM CPU outperforms x86 leaders. Lombardy hikes data centre fees 200%. Illinois mandates AI safety audits unanimously. Robinhood opens trading to autonomous agents.

imec fabricates the first quantum qubit on a High-NA EUV chip line. Eli Lilly’s gene therapy cuts cholesterol 62% in a single dose. China locks down AI talent travel. SpaceX sells orbital AI compute. Massachusetts rideshare drivers win first-in-nation union rights.

Pope Leo XIV makes AI ethics binding doctrine in a 42,300-word encyclical. The ECB summons 111 banks over Mythos vulnerabilities. Huawei unveils an EUV-free chip architecture. McKinsey clients demand outcome-based fees. Wind and solar surpass gas globally for the first time.

Anthropic’s Glasswing finds 10,000 software vulnerabilities — only 97 patched. DeepMind solves open Erdos problems at hundreds of dollars each. NTSB suspends its crash database after AI recreates dead pilots’ voices from a PDF. France commits another billion to quantum. ECB rejects euro stablecoins.

AI-designed materials target PFAS removal from drinking water in an industry first. ClickUp cuts 22% of staff and offers $1M for agent orchestrators. Memory hits 63% of AI chip costs. China activates a 10,000-unit humanoid production line. Retatrutide matches gastric bypass outcomes.

OpenAI disproves an 80-year-old math conjecture with original ideas. Nvidia posts $81.6B in quarterly revenue, then concedes China to Huawei. Anthropic projects its first profit at $10.9B. SpaceX files for IPO claiming a $28.5T market. Intuit cuts 17% of staff.

Google ships AI agents to 900 million users at I/O. Standard Chartered and Meta cut 15,000 jobs in one day. Blackstone and Google launch a $25 billion TPU venture. The Pope takes on AI in his first encyclical. Figure’s robots sort 180,000 packages nonstop for a week.

Bug bounty programmes drown in AI-generated reports as Anthropic briefs central banks on real vulnerabilities. Iran threatens subsea cables. NextEra buys Dominion for $67 billion. Chinese labs overtake US in video generation. Birth-rate plunge tracks smartphone adoption.

AI-assisted exploit cracks Apple’s M5 security in five days. Data centres cost $38 billion per gigawatt — and 49,000 Lake Tahoe residents may lose power to feed them. Citadel’s Ken Griffin calls AI productivity ‘profoundly more powerful.’ Malta gives every citizen ChatGPT Plus.

OpenAI fuses ChatGPT and Codex into one agent layer. Samsung’s 61,000 workers vote to strike over AI memory pay gaps. US employment in 18 AI-exposed roles falls for a second year. EY pulls a hallucination-riddled report. AI models now exploit Chrome’s V8 engine autonomously.

AI systems now optimise their own training and beat humans doing it. Palo Alto Networks warns of a three-to-five-month cybersecurity window. Cerebras IPOs at $95 billion. Anthropic hits $900 billion valuation. arXiv bans authors for unsupervised AI output.

UK Safety Institute finds AI capability doubling every four months with no observed ceiling. Recursive Superintelligence launches with $650M for self-improving AI. Claude for Small Business wires agents into payroll and invoicing. Three robotics deployment milestones in one day. Poland advances digital services tax despite US threats.

Anthropic ships Claude for legal with 20 connectors. Google embeds Gemini into the OS pointer. Princeton ends 133 years of unproctored exams over AI cheating. GPT-5.5 scores IQ 136. xAI installs 19 gas turbines to bypass the grid. Google and SpaceX explore data centres in orbit.

First AI-developed zero-day exploit confirmed in the wild. Thinking Machines ships real-time multimodal AI. GPT-5.5 finds fatal errors in a third of its own benchmark. OpenAI launches $4B enterprise unit. Cerebras IPO targets $49B. Amazon employees game AI adoption metrics with fake-task tool.

AI data centres hit walls in Maryland, Kenya, and Iraq. OpenAI shuts down fine-tuning. AMD closes 75x of the gap on Nvidia in two weeks. Women hold 83% of the most AI-vulnerable jobs. Experian reports 40% of breaches are now AI-powered.

METR can barely measure Claude Mythos autonomy. A Fields Medal winner reports PhD-level math in 17 minutes. Three weeks of AI pen testing matches a year of manual work. Cloudflare cuts 20% at record revenue. Apple taps Intel to fab its chips.

Anthropic takes over SpaceX’s Colossus with 220,000 GPUs after 80x Q1 growth. xAI dissolved into SpaceXAI. Motherboard sales collapse 25% as wafers redirect to AI. EU drafts cloud sovereignty rules. Chrome silently installs 4 GB of Gemini on every desktop.

GPT-5.5 halves hallucinations on high-stakes prompts. Apple lets users swap AI models like browsers — and pays $250M for overpromising. Samsung crosses $1 trillion. OpenAI commits $50B to compute this year. Coinbase cuts 14% as engineers ship in days what teams shipped in weeks.

White House reverses AI hands-off doctrine, proposing FDA-style model reviews. Anthropic and OpenAI lock $11.5B in PE joint ventures. Nvidia reports zero percent China market share. A battery-powered camera screens 15 million patients. Robots build clay homes from dirt.

UAE mandates 50% of government on agentic AI within two years. Hyperscaler capex hits $805B in 2026, $1.1T next year. Harmonic’s AI produces original math proofs experts call elegant. Boston Dynamics scales humanoids from four per month toward tens of thousands. South Africa withdraws AI-written AI policy.

Pentagon clears seven AI companies for classified networks. Iranian drone strikes leave Amazon cloud customers facing months of disruption. Chinese courts rule employers cannot fire workers to replace them with AI. Cerebras targets $40B IPO while OpenAI’s CFO pushes for delay.

NSA tests Anthropic’s Mythos for cyber vulnerability hunting. Huawei takes China’s AI chip crown as Intel posts its best month in 55 years. 1X targets 100,000 humanoids by 2027. Harvard finds AI outperforms physicians on diagnosis. Musk admits xAI distilled OpenAI’s models.

White House drafts guidance to bypass its own Anthropic ban for Mythos. Alphabet raises capex guidance to $190B. Nature study: friendlier AI models give worse answers by 10–30 points. Figure ships one humanoid per hour. Mayo Clinic AI detects pancreatic cancer three years early.

GPT-5.5 doubles on hard reasoning and writes its own GPU kernels. OpenAI goes multi-cloud while missing revenue targets. Google signs classified Pentagon AI deal. Tech layoffs hit 45,800 in March. AI discovers a pre-CRISPR gene-editing system. A third of new websites are AI-generated.