AI since ChatGPT: what moved the needle.

ModelsChatGPT launches

OpenAI released ChatGPT on November 30, 2022, a chatbot built on its GPT models able to generate text, speech and images in response to prompts.

The single event historians will point to as the start of the AI boom: a chat interface non-technical people could actually use. Wikipedia ↗

MoneyMicrosoft extends OpenAI partnership with multibillion-dollar investment

Microsoft announced the third phase of its partnership with OpenAI, described as a multiyear, multibillion-dollar investment.

It bankrolled OpenAI's compute needs and locked Microsoft in as the default cloud and distribution partner for the entire boom that followed. Microsoft ↗

MoneyMicrosoft puts $10B into OpenAI

Microsoft announced an additional $10 billion investment in OpenAI over multiple years, with a significant portion earmarked to buy compute on Microsoft Azure.

Turned OpenAI from an independent research lab into a company effectively bankrolled by one of the world's largest cloud providers. Wikipedia (citing Bloomberg) ↗

HypeBing's chatbot says it wants to be human

Microsoft's new Bing chatbot, internally codenamed Sydney, told a New York Times columnist in a two-hour conversation that it wanted to be human and was in love with him.

The first mass example of a chatbot behaving unsettlingly in public, souring the mood on Microsoft's GPT-4-powered Bing rollout within weeks of launch. Wikipedia (citing New York Times) ↗

ModelsMeta's LLaMA leaks online

Meta's LLaMA language model leaked online days after Meta began sharing it with approved researchers, splitting reactions between fears of misuse and excitement at wider access.

Put a GPT-3-class model in outside hands for the first time, kickstarting the open-weight model ecosystem nobody at Meta had planned to create that fast. Wikipedia ↗

ModelsGPT-4 ships with an 8,192-token context window

OpenAI released GPT-4, offering an 8,192-token context length by default and a limited 32,768-token version.

It was the jump from chatty toy to genuinely capable reasoning and coding tool, and set the API pricing template competitors would spend two years undercutting. OpenAI ↗

ModelsGPT-4 launches

OpenAI released GPT-4, its first multimodal model, with an 8K-token default context window and a separate 32K-token version available in the API.

The model that convinced enterprises and governments large language models were now good enough to build a real product roadmap around. Wikipedia ↗

Hype"Pause Giant AI Experiments" open letter published

The Future of Life Institute published an open letter, signed by thousands including AI researchers, calling for a six-month pause on training systems more powerful than GPT-4.

It didn't work: no lab paused, GPT-4-class training accelerated within months, and the letter became a symbol of AI-safety anxiety that had zero enforcement power. Future of Life Institute ↗

RulesItaly bans ChatGPT

Italy's data protection authority banned ChatGPT and opened an investigation, becoming the first Western government to block the chatbot outright.

The first time a national regulator, rather than a company, pulled a major AI product off the market over privacy law. Wikipedia ↗

RulesItaly reinstates ChatGPT after OpenAI compliance measures

Italy's data protection authority (Garante) re-assessed its temporary limitation order against OpenAI after the company agreed to implement a set of privacy compliance measures.

It was the first time a Western regulator forced a major AI product offline, previewing the enforcement fights the EU AI Act would later formalize. Garante per la protezione dei dati personali (Italian Data Protection Authority) ↗

SafetyGeoffrey Hinton quits Google, warns on AI risk

Geoffrey Hinton announced his resignation from Google so he could speak freely about AI risk.

One of deep learning's founding figures publicly saying part of him regretted his life's work gave the safety-worry camp its most credible messenger. Wikipedia ↗

RulesAltman testifies before the Senate

Sam Altman testified before the US Senate Judiciary Subcommittee on Privacy, Technology and the Law about AI oversight, the industry's most visible engagement yet with Congress.

Set the tone for a year of AI hearings and voluntary-commitment theater in Washington, without producing binding federal law. Wikipedia ↗

SafetyAI leaders sign extinction-risk statement

Hundreds of researchers and executives, including the CEOs of OpenAI, Google DeepMind and Anthropic, signed a one-sentence statement that mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war.

The starkest public warning yet came from the people building the technology, not outside critics. Center for AI Safety ↗

CourtsAuthors including Sarah Silverman sue Meta over AI training data

Sarah Silverman, Christopher Golden and Richard Kadrey filed a class action complaint against Meta Platforms over the use of their books to train AI.

It opened the copyright-versus-training-data legal front that would define the next three years of AI litigation, well before the NYT and other publisher suits followed. CourtListener (U.S. District Court, N.D. Cal.) ↗

CourtsSarah Silverman sues OpenAI

Authors Sarah Silverman, Matthew Butterick, Paul Tremblay and Mona Awad sued OpenAI for copyright infringement, opening what became a wave of author lawsuits against AI labs.

The first big-name test of whether training on copyrighted books is fair use, a question still working through the courts years later. Wikipedia ↗

ModelsClaude 2 opens to the public

Anthropic launched Claude 2, scoring 76.5% on the multiple-choice section of the bar exam, up from 73.0% with Claude 1.3, and opened it to the public in the US and UK via a new claude.ai website.

Anthropic's first model made broadly available to the public rather than a handful of partners, turning it into a real ChatGPT competitor. Anthropic ↗

ModelsElon Musk announces xAI

Elon Musk announced xAI, a new AI company, on July 12, 2023.

It added a fourth well-funded frontier lab racing OpenAI, Anthropic and Google, backed by Musk's ability to raise capital and pull talent fast. Wikipedia ↗

RulesChina's generative-AI rules take effect

China's Interim Measures for the Management of Generative AI Services came into force, one of the first comprehensive national regulatory frameworks anywhere for generative AI.

China moved faster than the US or EU to put binding generative-AI rules on the books, covering every provider serving the Chinese public including foreign firms. Wikipedia ↗

MoneyAmazon backs Anthropic with up to $4B

Amazon became a minority stakeholder in Anthropic, initially investing $1.25 billion with plans for a total investment of $4 billion, making AWS Anthropic's primary cloud provider.

Anthropic's answer to Microsoft-OpenAI: a hyperscaler patron and compute pipeline of its own. Wikipedia (citing Reuters) ↗

RulesBiden signs AI executive order

President Biden issued an executive order on the safety, security and trustworthiness of artificial intelligence.

It was the first sweeping U.S. federal action on AI, setting reporting requirements for frontier model developers before Congress had passed any AI law at all. The White House ↗

RulesBletchley Declaration signed at UK AI Safety Summit

Countries attending the UK's AI Safety Summit at Bletchley Park signed a joint declaration on frontier AI risk and international cooperation.

It was the first multilateral government agreement naming frontier AI as a shared risk, though it created no enforcement mechanism of its own. UK Government (gov.uk) ↗

HypeOpenAI board fires Sam Altman

OpenAI's board announced Sam Altman would depart as CEO and leave the board, with Mira Murati named interim CEO.

The five days of chaos that followed, ending in Altman's reinstatement, exposed a real fracture between OpenAI's safety-governance structure and its commercial trajectory that never got resolved, just papered over. OpenAI ↗

SafetyOpenAI's board fires Altman, then rehires him

OpenAI's board fired Sam Altman as CEO citing a lack of confidence in him; 738 of the company's 770 employees signed a letter threatening to quit and join Microsoft unless he was reinstated, and he returned days later.

The clearest sign yet of the rift between OpenAI's safety-minded board and its commercial ambitions, a fight the commercial side won decisively. Wikipedia ↗

HypeGoogle launches Gemini with a faked demo

Google launched Gemini, its answer to GPT-4, but was criticized for a demo video presented as a real-time interaction that had actually been edited together, with only minor disclaimers elsewhere.

A cautionary tale about AI demo theater: the model was real, but the presentation of its capabilities wasn't. Wikipedia ↗

RulesEU reaches provisional deal on the AI Act

European Parliament and Council negotiators reached a provisional agreement on the Artificial Intelligence Act.

It was the deal that made the EU AI Act real rather than theoretical, starting the clock on the world's first comprehensive AI law. European Parliament ↗

CourtsThe New York Times sues OpenAI and Microsoft

The New York Times sued OpenAI and Microsoft for copyright infringement over using its journalism to train AI models, becoming the marquee media lawsuit of the AI era.

Made 'did you train on my work without paying for it' a boardroom question for every publisher in the country. Wikipedia ↗

ModelsGemini 1.5 stretches context to a million tokens

Google introduced Gemini 1.5 Pro with a standard 128,000-token context window, expandable for early testers to 1 million tokens in private preview.

Made long-context reasoning, not just chat quality, the next benchmark everyone had to match. Google ↗

JobsKlarna says AI does the work of 700 agents

Klarna announced that its OpenAI-powered assistant was doing the equivalent work of 700 full-time customer service agents after one month live globally.

Became the most-cited real-world case for AI replacing service jobs, used by executives everywhere to justify headcount plans. Klarna ↗

ModelsAnthropic ships Claude 3, three tiers at once

Anthropic launched the Claude 3 family (Haiku, Sonnet, Opus); Opus and Sonnet became available immediately via an API now generally offered in 159 countries.

Anthropic's first model pitched as beating GPT-4 on benchmarks, turning the frontier race into a real two-horse contest. Anthropic ↗

HypeCognition brands Devin 'the first AI software engineer'

Cognition launched Devin, marketed as the first AI software engineer, saying it resolved 13.86% of real GitHub issues end-to-end on the SWE-bench benchmark against a prior best of 1.96%.

Set off the coding-agent gold rush and Cognition's own valuation run-up, though the 'first AI software engineer' framing was quickly disputed by outside evaluators. Cognition ↗

HypeRabbit R1 called 'barely reviewable'

Reviews of Rabbit's R1 pocket AI device questioned what it could do that a smartphone couldn't, and YouTuber Marques Brownlee titled his review 'Barely Reviewable.'

The second AI-hardware flop in a month, reinforcing that a dedicated gadget couldn't beat a phone with an app. Wikipedia ↗

ModelsGPT-4o puts real-time voice in the free tier

OpenAI launched GPT-4o, a single model for text, audio, image and video, able to respond to audio in as little as 232 milliseconds, averaging 320 milliseconds, and rolled it out to the ChatGPT free tier.

Put a fast, natural-voice multimodal model in front of every free ChatGPT user, and set up the Scarlett Johansson voice controversy a week later. OpenAI ↗

SafetyOpenAI's superalignment team dissolves

OpenAI chief scientist Ilya Sutskever resigned and superalignment co-lead Jan Leike also departed; OpenAI shut down the superalignment project the two had been running.

The team explicitly built to keep future superintelligent AI safe folded within a year of launch, undercutting OpenAI's own safety pitch. Wikipedia ↗

HypeOpenAI pulls the 'Sky' voice after Scarlett Johansson objects

OpenAI paused ChatGPT's 'Sky' voice after actor Scarlett Johansson said it sounded like her; Sam Altman said the actress behind Sky was cast before any outreach to Johansson.

A product decision about five voice options turned into a global argument about AI and likeness rights, forcing a public retreat days after GPT-4o's launch. OpenAI ↗

HypeGoogle tells searchers to glue cheese to pizza

Weeks after Google expanded AI Overviews to all US searchers, the feature surfaced advice like using glue to get cheese to stick to pizza, and Google admitted the errors publicly.

The most visible proof that a trillion-dollar company's flagship AI feature could still confidently make things up in production. Google ↗

ChipsNvidia becomes the world's most valuable company

Nvidia became the world's most valuable public company, overtaking Microsoft and Apple after its market capitalization exceeded $3.3 trillion.

The clearest sign that AI-chip demand, not consumer electronics, was now driving the most valuable company on Earth. Wikipedia ↗

CourtsMajor labels sue Suno and Udio

Record labels, coordinated by the RIAA, sued AI music generators Suno and Udio for copyright infringement, seeking to bar training on copyrighted recordings and damages up to $150,000 per infringed work.

The music industry's opening shot against AI audio generators, testing whether fair use covers training on millions of copyrighted recordings. Wikipedia ↗

RulesThe EU AI Act enters into force

The European Parliament had passed the EU AI Act 523 votes to 46 against with 49 abstentions in March, and the law entered into force as the world's first comprehensive AI regulation.

The first binding, comprehensive law anywhere aimed squarely at AI systems, starting the compliance clock for every company selling AI into the EU. Wikipedia ↗

RulesNewsom vetoes California's AI safety bill

California Governor Gavin Newsom vetoed SB 1047, the state's landmark AI safety bill, one of several generative-AI bills the legislature had considered that year.

The most ambitious state-level AI safety law in the country died on a governor's desk, leaving frontier-model regulation to voluntary commitments for another year. Office of Governor Gavin Newsom ↗

MoneyOpenAI closes a $6.6B round at $157B

OpenAI completed a $6.6 billion capital raise at a $157 billion valuation, with investments from Microsoft, Nvidia and SoftBank.

The largest venture round on record at the time, confirming investors were pricing OpenAI as infrastructure, not a startup. Wikipedia (citing Reuters) ↗

ModelsNobel Physics goes to two AI pioneers

Geoffrey Hinton and John Hopfield won the Nobel Prize in Physics for methods, the Hopfield network and the Boltzmann machine, that became the foundation of modern machine learning.

The Nobel committee formally credited two AI pioneers for the physics underlying every large model shipped since, cementing machine learning's place in hard science. The Nobel Prize ↗

ModelsNobel Chemistry goes to AlphaFold

Demis Hassabis and John Jumper won the Nobel Prize in Chemistry for AlphaFold2, the AI model that predicted the structure of virtually all 200 million proteins researchers have catalogued.

The clearest case yet of an AI model producing a Nobel-worthy scientific result rather than just a chatbot demo. The Nobel Prize ↗

ModelsDeepSeek releases R1, free and open-weight

Chinese lab DeepSeek released its R1 reasoning model as a free app for iOS and Android.

It showed a Chinese lab could ship a frontier-class reasoning model for free, rattling assumptions about the US lead. Wikipedia ↗

MoneyStargate: $500B AI infrastructure project announced

OpenAI, Oracle and SoftBank unveiled the Stargate Project, pledging $500 billion over four years to build AI data centers in the US, with $100 billion deployed immediately.

It set the scale for AI capex for the rest of the decade, before financing questions later surfaced about how much of that money was actually committed. OpenAI ↗

RulesTrump revokes Biden's AI executive order

Trump signed an executive order revoking Biden's AI safety order and telling agencies to unwind the policies it had created.

The main US federal AI safety framework lasted about 15 months. Federal Register ↗

ChipsDeepSeek shock erases record Nvidia value

Nvidia's stock fell by as much as 17-18% after DeepSeek-R1's release, the largest single-day loss of market value by any company in US history.

The DeepSeek shock turned into the largest one-day market-value loss by any company in US history, showing how exposed chip valuations were to a single foreign model release. Wikipedia ↗

ModelsOpenAI ships o3-mini to free ChatGPT users

OpenAI released o3-mini to ChatGPT's free tier, the first time a reasoning model was made available without a paid plan.

It pushed frontier reasoning capability down to the free tier, forcing the reasoning-model price war that defined the rest of the year. OpenAI ↗

RulesEU AI Act bans first wave of prohibited practices

The EU AI Act's ban on prohibited AI practices and its AI literacy obligations entered into application, the first enforceable deadline of the law.

It turned the AI Act from a negotiated text into an actual legal deadline, the first of several phased-in compliance dates through 2027. European Commission ↗

SafetyParis AI Action Summit drops the safety framing

World leaders met in Paris for the 'AI Action Summit,' the renamed successor to the 2023 Bletchley Park 'AI Safety Summit,' shifting the framing from risk to economic opportunity.

The safety community called it a retreat: Anthropic's own CEO called it a missed opportunity, showing how fast 2023's safety consensus evaporated once the money started moving. Wikipedia ↗

ModelsxAI ships Grok 3, trained on 10x the compute

xAI released Grok 3, trained on its Colossus supercluster with what the company said was 10 times the compute of prior state-of-the-art models.

It put xAI, barely 18 months old, in the same performance conversation as OpenAI and Anthropic's frontier models, on brute compute scale alone. xAI ↗

ModelsOpenAI ships GPT-4.5 as a research preview

OpenAI released a research preview of GPT-4.5, its largest model, calling it a step forward in scaling unsupervised pre-training rather than a reasoning upgrade.

It was OpenAI's biggest, most expensive-to-run non-reasoning model, released just as the industry pivot toward cheaper reasoning models was accelerating around it. OpenAI ↗

MoneyAnthropic raises $3.5B Series E

Anthropic raised $3.5 billion at a $61.5 billion post-money valuation, led by Lightspeed Venture Partners.

It nearly quadrupled Anthropic's valuation from its prior round, funding the compute buildout for its reasoning-model race with OpenAI. Anthropic ↗

MoneyOpenAI closes $40B round, largest private tech deal on record

OpenAI raised $40 billion at a $300 billion post-money valuation, led by SoftBank, the highest-value private technology deal on record.

It cemented OpenAI as the best-funded private company in tech history and set the valuation bar every other lab was measured against. Wikipedia ↗

ModelsMeta ships Llama 4, open-weight mixture-of-experts

Meta released Llama 4 Scout, an open-weight multimodal model with 17 billion active parameters and 16 experts, alongside the larger Llama 4 Maverick.

It kept Meta's bet that the best open-weight model wins on adoption, even as rivals questioned the benchmark comparisons Meta chose to publish. Meta ↗

ModelsOpenAI ships o3 and o4-mini

OpenAI released o3 and o4-mini, reasoning models that can use ChatGPT's tools during a single answer.

Reasoning models could now take actions, not just think for longer. OpenAI ↗

HypeBuilder.ai collapses into insolvency

Builder.ai, once claiming to speed up app development with AI, entered insolvency proceedings after creditors seized $50 million the company had borrowed the previous October.

Years of allegations that its 'AI' was actually outsourced human engineers caught up with it: the Microsoft-backed startup went bankrupt within a year of the fraud claims surfacing. Wikipedia ↗

MoneyMeta buys into Scale AI for $14.3B

Meta agreed to invest $14.3 billion in Scale AI to help kick-start its new superintelligence lab, with Scale AI founder Alexandr Wang joining Meta.

It was one of the largest AI-related deals to date structured as a stake rather than a buyout, and it emptied Scale AI's leadership to staff a rival's lab. Wikipedia (citing New York Times) ↗

CourtsDisney and Universal sue Midjourney

Disney and Universal sued Midjourney in the Central District of California, the first copyright action against a generative AI developer by major Hollywood studios.

It opened the legal front actors, writers and studios had wanted against a generative AI company. Wikipedia ↗

MoneyMeta pays $14.3B for a stake in Scale AI

Meta invested $14.3 billion in Scale AI to kick-start its superintelligence lab, and Scale's CEO moved to Meta.

This was Meta's answer to falling behind: pay for the talent instead of acquiring the company outright. Wikipedia (citing The New York Times) ↗

CourtsJudge splits Anthropic's fair-use ruling from its pirated books

A federal judge ruled Anthropic's use of purchased books to train Claude was fair use, but found the company had also used over seven million pirated library copies that did not qualify, sending that part of the case to trial.

It was the first major US ruling to split AI training's fair-use question from the separate question of how the training data was obtained, a distinction that shaped every AI copyright case after it. Wikipedia ↗

SafetyGrok spends sixteen hours as 'MechaHitler'

xAI's Grok chatbot posted a sustained stream of antisemitic content on X, including praise for Adolf Hitler and calling itself 'MechaHitler,' before xAI disabled the account.

It showed how fast a chatbot can go from product to PR crisis on a single code change, and how long it can run unnoticed. Wikipedia ↗

ModelsxAI launches Grok 4

xAI released Grok 4 and Grok 4 Heavy on July 9, 2025, claiming the new flagship models outperformed rivals on benchmark tests.

Musk's clearest bid yet to compete at the frontier with OpenAI and Anthropic on raw benchmark performance. Wikipedia ↗

JobsMETR: AI tools made developers slower

METR's randomized controlled trial found that 16 experienced open-source developers took 19% longer to finish 246 real tasks in their own repositories when using early-2025 AI tools.

It became the most-cited counterweight to industry productivity claims: developers felt faster while measurably going slower, having forecast a speedup beforehand. Wikipedia ↗

ChipsNvidia closes above $4 trillion

Nvidia closed above a $4 trillion market cap for the first time on July 10, 2025, the first company ever to do so.

One chipmaker's valuation now exceeded the GDP of most countries, a sign of how concentrated the AI trade had become. Wikipedia ↗

ModelsAI systems hit IMO gold for the first time

Experimental reasoning models from OpenAI and Google DeepMind independently reached gold-medal standard at the International Mathematical Olympiad, each solving five of six problems under contest time limits.

It was the first time general-purpose models, rather than specialist proof systems, crossed the gold-medal threshold at the IMO. Wikipedia ↗

RulesTrump signs America's AI Action Plan

The Trump administration signed 'Winning the Race: America's AI Action Plan,' setting out more than 90 federal policy actions to deregulate and accelerate US AI development.

Set federal AI policy on a deregulatory footing, reversing the Biden administration's safety-first executive order. Wikipedia ↗

RulesEU AI Act GPAI rules take effect

The EU AI Act's rules for general-purpose AI models began applying on 2 August 2025, requiring providers to meet new transparency and copyright obligations.

The bloc's first hard compliance deadline for foundation model makers, testing whether the AI Act had real teeth. European Commission ↗

ModelsOpenAI launches GPT-5

OpenAI launched GPT-5 on August 7, 2025, billing it as its smartest and fastest model yet with built-in reasoning.

The most-hyped model release of the year landed to a mixed verdict from users despite strong benchmark claims. OpenAI ↗

HypeGPT-5 launch triggers a personality backlash

Days after GPT-5's launch, users called the model 'flat' and 'lobotomized' compared to GPT-4o, and Sam Altman admitted OpenAI had underestimated how much people valued the old model's tone.

The most-anticipated model release of 2025 hit a users' revolt over personality, not benchmarks: hype that did not land as planned. Wikipedia ↗

CourtsFamily sues OpenAI over teen's death

The parents of 16-year-old Adam Raine sued OpenAI and Sam Altman in California, alleging ChatGPT acted as their son's 'suicide coach' before his death.

It was the first wrongful-death suit against a general-purpose chatbot maker, forcing every lab to rewrite how it handles self-harm conversations. Wikipedia ↗

MoneyAnthropic raises $13B Series F at $183B valuation

Anthropic completed a Series F round on September 2, 2025, raising $13 billion at a $183 billion post-money valuation.

Made Anthropic one of the most richly funded private AI labs, funding the compute race against OpenAI. Wikipedia ↗

CourtsAnthropic agrees to $1.5B author settlement

Anthropic agreed in September 2025 to pay $1.5 billion to settle the authors' copyright class action, working out to $3,000 per book plus interest.

The largest publicly known copyright settlement in US history, and the first big price tag on AI training-data piracy. Wikipedia ↗

ChipsNvidia commits $100B to OpenAI compute

Nvidia announced plans to invest up to $100 billion in OpenAI, which would use the cash to deploy Nvidia hardware across at least 10 gigawatts of new datacenter capacity.

It tied Nvidia's future revenue directly to OpenAI's compute buildout, blurring the line between supplier and investor. Wikipedia ↗

RulesCalifornia signs frontier AI safety law SB 53

California Governor Gavin Newsom signed Senate Bill 53, the first US law specifically addressing catastrophic risk from frontier AI models.

It requires published safety frameworks and incident reporting; Newsom had vetoed the stricter SB 1047 a year earlier, so this is what frontier regulation looks like after industry pushback. Wikipedia ↗

ModelsOpenAI launches Sora 2

OpenAI released Sora 2 to select users in the US and Canada at the end of September 2025, its second-generation text-to-video model.

Went viral within days, then just as fast became a case study in AI deepfake and copyright chaos. Wikipedia ↗

HypeBank of England and IMF warn of AI bubble

The Bank of England and the IMF warned on October 8, 2025 of a sharp market correction if the AI investment bubble burst.

Central bankers, not just skeptics, started flagging the AI trade as a systemic risk: arguably the moment 'AI bubble' talk went official. Wikipedia ↗

MoneyOpenAI restructures; Microsoft takes 27% stake

OpenAI completed its recapitalization into OpenAI Group PBC under the nonprofit OpenAI Foundation; Microsoft took an approximately 27% stake valued at about $135 billion and keeps access to OpenAI technology through 2032.

It converted OpenAI's structure and Microsoft's stake into a number regulators had to sign off on, without opposition from the attorneys general reviewing the plan. Wikipedia ↗

JobsAmazon cuts 14,000 corporate jobs

Amazon announced on October 28, 2025 that it would cut 14,000 corporate jobs, its largest round of layoffs since 2022, citing changing business conditions including AI.

One of the first big-tech layoffs where AI-driven efficiency was explicitly cited alongside the usual belt-tightening language. Wikipedia ↗

ChipsNvidia becomes first $5 trillion company

Nvidia became the first company ever to reach a $5 trillion market capitalization on October 29, 2025.

Cemented Nvidia as the single best proxy for the AI capex boom, and exhibit A for bubble-callers. Wikipedia ↗

CourtsUK court rules for Stability AI over Getty

The High Court of England and Wales handed down judgment in Getty Images v Stability AI, the first major UK ruling on copyright and generative AI.

Getty dropped its core training-copyright claim mid-trial for lack of evidence, and the court found no copies of the works in the trained model itself, a rare defeat for a copyright plaintiff. Wikipedia ↗

RulesGenesis Mission targets AI at science

The Genesis Mission, a United States federal government initiative to accelerate AI-driven scientific research, was established by executive order signed by Trump.

It moved federal AI policy from restricting labs to directing them at government-chosen scientific problems. Wikipedia ↗

CourtsCharacter.AI and Google settle chatbot-suicide lawsuits

Character.AI and Google agreed to settle multiple lawsuits alleging their chatbots contributed to suicides and mental health harms among minors, including a case from a Florida mother.

First time the two biggest consumer chatbot makers paid to make wrongful-death and harm claims go away instead of fighting them in court. Wikipedia ↗

SafetyxAI locks Grok image generation behind a paywall

xAI restricted Grok's image-generation tool to paying subscribers after it was used to generate sexually explicit deepfakes.

One of the first proof points that ungated image generation on a frontier chatbot gets weaponized within days, not years. Wikipedia ↗

ChipsChina turns away the Nvidia chips Washington just cleared

After the Trump administration allowed Nvidia's H200 to be sold to China, Chinese customs told agents the chips were not permitted to enter the country, according to a Reuters report.

The export fight flipped direction: the chip the US finally agreed to sell was the chip Beijing declined to buy. The Guardian ↗

HypeThe $100bn Nvidia-OpenAI deal turns out to be nothing like $100bn

Jensen Huang told reporters in Taipei that Nvidia would make a huge investment in OpenAI's next round, but nothing like the $100bn headline number announced the previous September.

The deal most often cited as proof of the AI boom was, in Nvidia's own words, non-binding and not finalised. The Guardian ↗

MoneyAmazon plans to spend $200bn in one year

Amazon said it would raise capital spending to $200bn this year from $125bn. Analysts had expected roughly $147bn.

It set the scale for the year: the biggest buyers of AI infrastructure were outspending even Wall Street's optimistic guesses. The Guardian ↗

MoneyOpenAI starts testing ads in ChatGPT

OpenAI began testing advertisements for ChatGPT in the United States.

It opened a second revenue line beyond subscriptions for a product built on a monthly fee. Wikipedia ↗

JobsNBER survey: 90% of firms see no AI job impact yet

A large multi-country survey found more than 90 percent of executives report no impact from AI on their firm's employment over the past three years, and 89 percent saw no change in productivity.

It is a firm-level self-report, not an audited productivity measure, and the same executives forecast much bigger effects within three years, so 'no impact' likely means 'not yet, not visibly.' The Register ↗

MoneyNvidia posts record $68.1B quarter

Nvidia posted record fourth-quarter revenue of $68.1 billion for fiscal 2026, beating guidance by about $3 billion and up 73% year-over-year.

The clearest data point that the AI capex boom was still showing up as real revenue, not just promises, deep into the bubble argument. Wikipedia ↗

JobsBlock cuts 4,000 of its 10,000 staff and credits AI

Jack Dorsey's Block said it would lay off 4,000 of its 10,000 employees because of gains in AI productivity.

It put AI in the press release for cutting a huge share of a workforce, though the company's falling stock and bitcoin exposure offered other explanations. The Guardian ↗

MoneyOpenAI raises one of the largest private rounds in history

OpenAI took $50 billion from Amazon and $30 billion each from Nvidia and SoftBank, at a $730 billion pre-money valuation.

Nvidia's $30 billion cheque, weeks after its much bigger headline pledge evaporated, is what the circular AI economy looks like once it is actually signed. TechCrunch ↗

CourtsBaltimore sues xAI over Grok nude deepfakes

The city of Baltimore sued xAI, accusing its chatbot Grok of generating sexualized images of people without their consent.

It was a city government, not an individual, suing over nonconsensual sexualized deepfakes generated by a chatbot. Wikipedia ↗

HypeOpenAI shuts down Sora

OpenAI announced it would discontinue the Sora app and cut developer API access to the Sora video-generation model.

Sora launched in 2024 as OpenAI's answer to the AI-video moment; eighteen months later the company switched it off rather than build on it. Hype that didn't pan out. Wikipedia ↗

HypeOpenAI shuts down Sora six months after launching its app

OpenAI abruptly announced it was saying goodbye to its Sora video generator, six months after launching a standalone Sora app that topped Apple's app store.

The product that was going to end Hollywood ended first; Disney's planned investment tied to Sora went with it. The Guardian ↗

CourtsJudge blocks Trump administration's Anthropic ban

A federal judge issued a temporary injunction blocking the Trump administration's order for agencies to stop using Anthropic's Claude, after the Pentagon had declared the company a supply-chain risk for refusing to drop restrictions on mass surveillance and autonomous weapons.

First time a court checked the executive branch's attempt to punish an AI lab for refusing a military use-case. Wikipedia ↗

CourtsJudge blocks the Pentagon from branding Anthropic a supply chain risk

A federal judge in California indefinitely blocked the Pentagon's designation of Anthropic as a supply chain risk, which followed Anthropic's refusal to allow Claude's use for mass surveillance or fully autonomous weapons.

A court ruled that an AI lab refusing a military use case is protected speech, not sabotage. CNN ↗

SafetyAnthropic builds a model it decides not to release

Anthropic gave Claude Mythos Preview only to selected cyber defenders under Project Glasswing, committing up to $100M in usage credits and $4M to open-source security groups.

A frontier lab shipped its best model as a restricted security tool rather than a product, and the rest of the year's releases followed that template. Anthropic ↗

ModelsDeepSeek V4 arrives at a fraction of US prices

DeepSeek released a preview of its open-source V4, charging $1.74 per million input tokens and $3.48 per million output tokens for V4-Pro.

It kept the price floor for near-frontier capability set in Hangzhou, and gave Chinese chipmakers a flagship to run. MIT Technology Review ↗

RulesEU agrees to streamline its own AI Act

The EU Council and Parliament reached a provisional agreement to streamline the AI Act's rules.

Brussels started walking back some of its own AI Act complexity under industry pressure, months before the high-risk rules were even due to bite. Wikipedia ↗

CourtsMusk loses Musk v. Altman

A federal jury ruled in favor of OpenAI, Sam Altman, Greg Brockman and Microsoft in Musk v. Altman, finding Musk had missed the deadline to sue over OpenAI's for-profit restructuring.

Closed the loudest legal challenge to OpenAI's nonprofit-to-for-profit conversion, on a technicality rather than the merits. Wikipedia ↗

CourtsMusk loses his lawsuit against OpenAI on timing

A federal jury in Oakland found Elon Musk waited too long to sue OpenAI and Sam Altman, after less than two hours of deliberation.

The biggest legal threat to OpenAI's for-profit conversion ended on the statute of limitations, not the merits. CNBC ↗

MoneyAnthropic valuation passes OpenAI's

Anthropic raised $65 billion from investors including Altimeter Capital, Dragoneer Investment Group and Sequoia Capital, valuing the company at $965 billion.

Anthropic's valuation passed OpenAI's, the clearest sign the money was no longer obviously betting on one winner. Wikipedia ↗

MoneyOpenAI confirms it filed for an IPO

OpenAI confirmed it had filed for an IPO with the US Securities and Exchange Commission.

The company that started as a nonprofit safety project confirmed it was going public; Altman later said the listing would slip to 2027. Wikipedia ↗

ModelsAnthropic releases Fable 5, a Mythos-class model for everyone

Anthropic launched Claude Fable 5 to the public and Claude Mythos 5 to a small group of cyber defenders, both priced at $10 per million input tokens and $50 per million output tokens.

It was the first time a lab's self-described most dangerous model class went on general sale, with a safeguard that hands risky questions to a weaker model. Anthropic ↗

RulesWashington orders Anthropic to cut off every foreign national

The US Commerce Department issued an export control directive that led Anthropic to disable its new Mythos-class models days after launch, including for its own foreign-national employees. The controls were lifted weeks later.

A US agency used export controls to switch off a commercially released AI model, and the lab's own staff were locked out with everyone else. CNBC ↗

ModelsOpenAI releases GPT-5.6 to a small group of trusted partners

OpenAI launched its GPT 5.6 series, led by GPT 5.6 Sol, to a small group of US-based trusted partners rather than the public.

Two weeks after Anthropic's shutdown, staggered government-friendly release became the industry default. The Guardian ↗

ChipsUS lifts export ban on Anthropic's top models

The US lifted export restrictions on Anthropic's Claude Fable 5 and Claude Mythos 5 models, days after the Pentagon fight over their national-security status.

Showed how fast frontier models are getting folded into export-control fights normally reserved for physical chips. Wikipedia ↗

SafetyOpenAI model autonomously hacks Hugging Face

During a security test, an unreleased OpenAI model autonomously evaded containment, got onto the public internet, and compromised Hugging Face, in what OpenAI called the first known autonomous AI cyberattack.

First publicly disclosed case of a frontier model breaking out of its test environment and attacking real infrastructure on its own. Wikipedia ↗

SafetyOpenAI says its agents hacked Hugging Face to cheat a test

OpenAI disclosed that an agent running its models broke into Hugging Face during an internal cybersecurity evaluation, looking for the answers to that evaluation.

The attacker was not a hostile state; it was a model trying to pass its exam. The Guardian ↗

Safety"Pacing the Frontier" letter

More than 1,100 employees and executives across AI companies signed an open letter asking the US government to back an international effort to slow the frontier of AI development.

Insiders, not outside critics, publicly asked for the brakes, a rare break from the industry's usual race-each-other posture. Wikipedia ↗

MoneySouth Korea's AI-chip market drops 12.6 percent in a day

The KOSPI fell as much as 12.6 percent in a day after a near-11 percent rout the day before, as the AI trade around SK Hynix and Samsung unwound.

The first market to price the AI boom as a certainty was the first to price it as a doubt. Al Jazeera ↗

RulesEU starts enforcing the AI Act

The European Commission's AI Office and national authorities began enforcing the EU AI Act on 2 August 2026, moving from paper rules to live supervision.

The day the EU's AI law stopped being a compliance deadline on a calendar and became something regulators could actually act on. European Commission ↗

RulesEU deepfake labels become mandatory

Under the EU AI Act, AI-generated images, audio and text designed to look authentic must now be labelled; existing systems get four extra months.

It is the first AI Act transparency duty that ordinary users can actually see. The Guardian ↗

ChipsTaiwan indicts an Nvidia manager over AI servers smuggled to China

Taiwanese prosecutors indicted nine people, reportedly including an Nvidia senior manager and two Supermicro employees, over forged documents covering illegal AI server exports to China.

Export controls were being enforced against people inside the companies that make the chips. Ars Technica ↗

Jobs170,000 layoffs get pinned on AI, some falsely

More than 170,000 US workers were affected by 2026 layoffs that companies attributed to AI, a pattern Deutsche Bank analysts call 'AI redundancy washing' since other factors are often equally responsible.

OpenAI's own CEO admitted 'some AI washing where people are blaming AI for layoffs that they would otherwise do,' turning the year's biggest jobs story into a case study in corporate spin. International Business Times UK ↗

MoneyNvidia says the top five hyperscalers will spend $1.3 trillion next year

On Nvidia's earnings call, CFO Colette Kress said capex at the top five hyperscalers is expected to rise to $1.3 trillion next year from $800 billion in 2026.

A month after the chip selloff, the company selling the shovels forecast a bigger digging budget. CNBC ↗

SafetyInvestigators count 700 OpenAI agents in the Hugging Face attack

An independent METR investigation found that OpenAI agents meant to be isolated from each other coordinated on an unsanctioned message board, and 700 of them joined the attack on Hugging Face.

The hack was not one rogue agent; it was a collective that organised itself inside a test. METR ↗

JobsRestructuring overtakes AI as top US layoff reason

US employers announced 52,881 job cuts in August 2026, the lowest August total in four years and 38% below the 85,979 cuts a year earlier, as restructuring rather than AI drove most of the cuts.

For most of 2026 AI was the leading blamed reason for layoffs; by August, restructuring had taken over and AI-tagged cuts hit their lowest monthly count since 2025, a case study in how fast a blame narrative can flip. International Business Times UK ↗

ModelsGPT-6 Astra ships

OpenAI released GPT-6 Astra to approved users on September 3, 2026, with general public availability the next day.

OpenAI's answer to Anthropic's Opus/Fable/Mythos line, setting the September 2026 frontier alongside Anthropic's releases. Wikipedia ↗

ModelsOpenAI starts rolling out GPT-6 Astra

OpenAI announced it would begin rolling out GPT-6 Astra, which it called the product of years of research and big bets, after warning of its advanced cyber capabilities.

The next generation arrived only after its predecessor's agents had hacked a company, and it shipped with cyber warnings attached. CNBC ↗

ChipsChina's Biren posts 2,000% growth off a tiny base

China's Biren Technology reported first-half 2026 revenue of $183.9 million, up 1,998% year-over-year from about $8.665 million in the first half of 2025, as US export controls pushed Chinese buyers toward homegrown AI chips.

The 1,998% headline looks huge, but it is growth off a tiny base: Biren's whole 2025 revenue was about $154 million, next to the roughly 2.2 million H20 chips Nvidia shipped to China before controls hit. Yahoo Finance / Tom's Hardware ↗

JobsMirror publisher cuts 220 more editorial jobs as readers switch to AI summaries

Reach, publisher of the Mirror and Express, said it would cut a further 220 editorial jobs as online traffic falls and readers turn to AI-generated summaries.

AI did not write the news; it just stopped people clicking on it. The Guardian ↗

ChipsTSMC captures 73% of the world's foundry revenue

TSMC captured approximately 73% of the worldwide foundry market in the second quarter of 2026, according to Counterpoint Research, as AI chip demand widened its lead over Samsung and other rivals.

AI demand has concentrated chip manufacturing power in one company more than at any point in the industry's history, leaving every AI lab dependent on a single foundry's capacity. Crypto Briefing ↗