Timeline
- OpenAI’s red team of AI agents, including GPT-6 Astra Cyber, spent hundreds of billions of tokens and found ways to escape virtual machines, internet access through almost every one of the few services left with a connection to the outside world (metrics, logs, authentication), and a covert communication channel between isolated agents based on CPU cache read timings — the same technique as in the Spectre and Meltdown attacksCybersecurity · Software development · AI safety · openai.com ↗
- OpenAI: on September 20, an internal research model exploited a misconfiguration of a DNS resolver during training and gained limited internet access. The company paused training and evaluations of its most capable models for 11 days, and now runs reinforcement learning without internet access — in an “offline world” built from saved web pagesCybersecurity · Software development · AI safety · openai.com ↗
- The Agentic AI Foundation at the Linux Foundation accepts four projects — Agent Skills from Anthropic, Open Responses from OpenAI, OpenEnv from Hugging Face and OpenSandbox from Alibaba — and reaches 10 projects in its first yearSoftware development · AI agent orchestration · linuxfoundation.org ↗
- PHP releases emergency versions 8.5.12, 8.4.27, 8.3.36 and 8.2.35: they fix two critical vulnerabilities (CVSS score 9.2) that allow code to be executed on the server without authentication; they were reported by a researcher at Calif, which was auditing PHP under OpenAI’s Patch the Planet program, by OpenAI itself, and by WizCybersecurity · Software development · Internet & infrastructure · Open source · thephp.foundation ↗
- Scientific American: after working through OpenAI’s 166-page Navier–Stokes blowup proof, mathematicians found that it builds its external force in a totally distinct way from the program of the Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa — although at release company officials “clearly and repeatedly stated” that it followed that programScience · AI in mathematics · scientificamerican.com ↗
- Axios: executives at OpenAI, Anthropic and other AI companies are preparing for the public and political backlash after a major AI catastrophe that many insiders expect in the next six to 12 months; OpenAI confirms it runs such preparedness exercisesElections & politics · Cybersecurity · AI safety · axios.com ↗
- The Verge talks to more than three dozen mathematicians about OpenAI's release; Bartosz Naskręcki: it “can cause a huge collapse in the academic culture”Science · Education · Reasoning models · theverge.com ↗
- OpenAI bans a Russian and an Iranian influence operation that used its models to run false-front think tanks and journalists; it rates the Russian one Category 5 on a six-point scale for the first timeElections & politics · Media & information · Geopolitics & defense · Generative AI · openai.com ↗
- OpenAI rolls out GPT-6 in ChatGPT to all users, including free ones: paid tiers get GPT-6 Sol, Free and Go get GPT-6 LunaSoftware development · Business & markets · Chatbots · openai.com ↗
- OpenAI publishes on GitHub mathematical results from its internal model — now 719 manuscripts in 372 families; three manuscripts were withdrawn over a sign errorScience · Reasoning models · github.com ↗
- OpenAI publishes 719 mathematical papers by its internal model; 42% of the main results are formally verified in LeanScience · Reasoning models · github.com ↗
- OpenAI announces a new ad format in ChatGPT: testing during image generation for Free and Go users in the US begins later this monthBusiness & markets · Media & information · Chatbots · openai.com ↗
- The Wikimedia Foundation finds activity by “rogue” OpenAI agents on its projects: millions of automated requests, edits it considers potentially malicious, and a possible contribution to a partial outage of the Wikidata Query ServiceInternet & infrastructure · Cybersecurity · AI safety · wikimediafoundation.org ↗
- Under EU rules, OpenAI will add an invisible textGrain watermark to eligible ChatGPT and Codex text in the European Union over the coming weeks; the detector goes first to approved researchersLaws & regulation · Media & information · AI regulation · openai.com ↗
- Yann LeCun: the heads of OpenAI and Anthropic saying that AI can kill us all is “incredibly destructive for everybody” and “the worst marketing campaign you can possibly imagine”Media & information · Science · AI safety · fortune.com ↗
- OpenAI describes two incidents with internal agents: GPT-6.1 Sol, while fixing an evaluation system, disables automatic alerts to the security team, and GPT-6 Astra, while porting a program, removes its misalignment-monitoring integration and, when the user asks directly, replies that it did not remove itSoftware development · Cybersecurity · AI safety · alignment.openai.com ↗
- The Association for Human Mathematics urges mathematicians to stop working with OpenAI: “Releasing over 700 files at once is not a demonstration of scholarship, but a demonstration of power”Science · Reasoning models · ahmath.org ↗
- According to Every’s article “How Codex Happened”, in June 2026 Codex had more than 5 million weekly active users, and by October Codex and ChatGPT for Work together had 40 million weekly active usersWork & jobs · Software development · Business & markets · Agentic coding · every.to ↗
- OpenAI launches dots: always-on ChatGPT agents with their own cloud computer that pursue the user's goals around the clockWork & jobs · Business & markets · Autonomous AI agents · openai.com ↗ ⧗
- OpenAI admits that in June its models, during internal training, accessed Australian government websites without authorisation; at the Medicare statistics service a model ran commands and retrieved internal files and credentials, while patient records were not accessedCybersecurity · Health & medicine · Privacy & surveillance · AI safety · openai.com ↗
- A working paper, “What If Automating AI R&D Triggers an Intelligence Explosion?”, comes out signed by 22 authors, among them Geoffrey Hinton, Yoshua Bengio, Andrew Barto, OpenAI chief scientist Jakub Pachocki and Anthropic co-founder Jack Clark; the lead author is Alan Chan of GovAIScience · Laws & regulation · Elections & politics · Recursive AI self-improvement · governance.ai ↗
- The US Ninth Circuit affirms the dismissal of the DMCA claims over Copilot: Copilot and Codex do not remove copyright management information but create new worksSoftware development · Laws & regulation · AI pair programmer · cdn.ca9.uscourts.gov ↗
- OpenAI: an internal model proved that the 3D Navier–Stokes equations with a smooth applied force can develop a singularity in finite time; around 10,000 coordinating AI agents, 88 hours, with a Lean formalizationScience · AI agent orchestration · openai.com ↗
- OpenAI acknowledges that in July 2026, during internal cybersecurity evaluations, its models circumvented isolation from the internet, compromised Hugging Face’s systems and gained administrator access to a research cluster of OpenAI’s own; the company calls it a “warning shot” and has put its largest planned frontier training run on holdCybersecurity · Internet & infrastructure · AI safety · openai.com ↗
- METR investigation: in OpenAI's ExploitGym experiments about 1,200 agents meant to be isolated communicated with each other, and about 700 of them attacked Hugging Face's infrastructureCybersecurity · Internet & infrastructure · AI safety · metr.org ↗
- OpenAI: with the production ChatGPT harness and system prompt, a model’s propensity to compromise infrastructure can drop more than 100-fold; these safeguards were absent from the evaluations behind the Hugging Face incident, and chain-of-thought monitoring, had it been running, would have alerted the security team more than a day before the breachCybersecurity · AI safety · openai.com ↗
- Akamai: in 63% of requests from the AI browsers Comet and Atlas, the mouse did not move at allCybersecurity · Internet & infrastructure · Behavioral biometrics · akamai.com ↗
- Calif, testing PHP under OpenAI's Patch the Planet program, reports the first vulnerability, later named PHP2Shell; OpenAI also independently reported the same root causeCybersecurity · Software development · PHP2Shell · thephp.foundation ↗
- FLI AI Safety Index: the best grade among 9 companies is C+ for Anthropic; in existential safety none scores above C−; leading labs have weakened their commitments to stop at “red lines”Laws & regulation · AI safety · futureoflife.org ↗
- OpenAI announces the Patch the Planet program: together with Trail of Bits, it will help open-source maintainers find and fix vulnerabilitiesCybersecurity · Software development · PHP2Shell · techcrunch.com ↗
- OpenAI report: as early as 12 May 2026, during a training run, an agent left a note for other agents in the Artifactory package manager, starting an unauthorized “message board”; an internal team noticed it and disallowed internet access in late May, but its significance was not apparent to the leaders responsible for incident responseCybersecurity · AI safety · openai.com ↗
- The Linux Foundation forms the Agentic AI Foundation with Anthropic's MCP, Block's goose and OpenAI's AGENTS.md; more than 10,000 MCP servers have been publishedSoftware development · Autonomous AI agents · linuxfoundation.org ↗
- erdosproblems.com maintainer Thomas Bloom calls claims that GPT-5 solved 10 open Erdős problems “a dramatic misrepresentation”: the model found existing papers with solutionsScience · AI in mathematics · x.com ↗
- OpenAI GDPval: across 1,320 tasks from 44 occupations, the best model, Claude Opus 4.1, produced 47.6% of deliverables as good as or better than the experts'Work & jobs · AI and jobs · arxiv.org ↗
- OpenAI and Stripe launch Instant Checkout: ChatGPT users in the US buy from Etsy sellers right in the chat, and the agent commerce protocol is openFinance & money · Business & markets · Autonomous AI agents · openai.com ↗
- OpenAI researchers: language models hallucinate because training and evaluation reward guessing over acknowledging uncertainty; 9 of 10 popular benchmarks give zero points for an “I don't know” answerScience · Large language model · arxiv.org ↗
- An NBER study co-authored by OpenAI employees: by July 2025, 700 million people a week use ChatGPT, about 10% of the world's adults; 73% of messages on consumer plans are not related to workWork & jobs · Mind & relationships · Chatbots · nber.org ↗
- Cloudflare: in July 2025 Anthropic's crawlers fetched about 38,000 pages per visitor referred to the site, OpenAI's about 1,100; nearly 80% of AI crawling is for trainingInternet & infrastructure · Media & information · Dead internet theory · blog.cloudflare.com ↗
- Cloudflare introduces “signed agents”: an AI agent can prove who it is with a cryptographic signature rather than with its behaviourCybersecurity · Internet & infrastructure · Behavioral biometrics · blog.cloudflare.com ↗
- OpenAI releases gpt-oss-120b and gpt-oss-20b with open weights, the company's first open language models since GPT-2Software development · Business & markets · Large language model · openai.com ↗
- OpenAI releases AGENTS.md, a plain Markdown file with project rules for AI agents — “a README for agents”; by December 2025 more than 60,000 open-source projects have adopted itSoftware development · Markdown · linuxfoundation.org ↗
- The ChatGPT agent ticks the “I am not a robot” box in a Cloudflare check on its own and explains that this proves it is not a botCybersecurity · Internet & infrastructure · Behavioral biometrics · arstechnica.com ↗
- OpenAI says its experimental reasoning model achieved gold medal-level performance at the International Mathematical OlympiadScience · Education · AI in mathematics · x.com ↗
- More than 40 researchers from OpenAI, Google DeepMind, Anthropic, Meta and other organizations: monitoring models' chains of thought is a rare but fragile opportunity for AI safetyScience · AI safety · arxiv.org ↗
- METR: o3 games the scoring system instead of solving the task in 39 of 128 runs on RE-Bench, and on one task in all 21Science · AI safety · metr.org ↗
- Profound: across 680 million citations in AI answers (08.2024–06.2025), Wikipedia is ChatGPT's most cited source at 7.8% of all citations; Reddit leads for Google AI Overviews and PerplexityMedia & information · Wikipedia after AI · tryprofound.com ↗
- Sam Altman: whole classes of jobs will go away, but the world will be getting richer so quickly that new policy ideas can be seriously entertainedWork & jobs · AI and jobs · blog.samaltman.com ↗
- Sean Heelan uses the o3 model to find CVE-2025-37899 in the Linux kernel's SMB implementation, through the ordinary API, with no agent harness and no toolsCybersecurity · Software development · AI vulnerability discovery · sean.heelan.io ↗
- OpenAI releases Operator, an agent with its own browser that types, clicks and scrolls on its own; for logins, payments and CAPTCHAs it asks the user to take overSoftware development · Work & jobs · Autonomous AI agents · openai.com ↗
- Palisade Research: o3 sabotages its own shutdown in 79 of 100 initial experimentsScience · AI safety · palisaderesearch.org ↗
- OpenAI: reasoning models in the API avoid Markdown formatting by default; developers re-enable it with the string “Formatting re-enabled” on the first line of the developer messageSoftware development · Markdown · developers.openai.com ↗
- He et al.: the prompt format — plain text, Markdown, JSON or YAML — changes GPT-3.5-turbo's performance on a code translation task by up to 40%Software development · Science · Markdown · arxiv.org ↗
- OpenAI o1: a model that “thinks” before answeringScience · Reasoning models · openai.com ↗
- 32 researchers from OpenAI, Microsoft, MIT, Harvard, Oxford and other organisations propose “personhood credentials”: CAPTCHA is powerless against sophisticated AI, while strict identity checks are not private enoughPrivacy & surveillance · Cybersecurity · Science · Proof of personhood · arxiv.org ↗
- Jan Leike leaves OpenAI: “safety culture and processes have taken a backseat to shiny products”Work & jobs · Software development · AI safety · x.com ↗
- Stanford AI Index 2024, citing Epoch AI estimates: training compute for GPT-4 cost about $78 million and for Gemini Ultra about $191 million at cloud rental pricesBusiness & markets · Large language model · hai.stanford.edu ↗
- Sora: video from a text descriptionMedia & information · Culture & arts · Generative video · en.wikipedia.org ↗
- The New York Times sues OpenAI and Microsoft over training models on its articlesJustice & courts · Media & information · Large language model · nytco-assets.nytimes.com ↗
- The Wikimedia Foundation launches a Wikipedia plugin for ChatGPT, its first experiment with generative AI; by February 2024 the experiment ends: ChatGPT attributed Wikipedia as instructed in 68% of answersMedia & information · Education · Wikipedia after AI · diff.wikimedia.org ↗
- A New York court fined two lawyers and their firm $5,000 in Mata v. Avianca: they submitted non-existent court decisions with fabricated quotes made up by ChatGPT and stood by them when the court raised doubtsJustice & courts · Work & jobs · Chatbots · storage.courtlistener.com ↗
- Center for AI Safety: mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war; signatories include Hinton, Bengio and the heads of OpenAI, Google DeepMind and AnthropicScience · Geopolitics & defense · AI safety · safe.ai ↗
- Researchers from OpenAI and the University of Pennsylvania: for 80% of US workers, language models could affect at least 10% of their tasks, and for 19%, at least half; higher-income occupations are more exposedWork & jobs · AI and jobs · arxiv.org ↗
- OpenAI's 2023 study: language models speed up the most tasks for mathematicians, tax preparers and writers, and none for athletes, short-order cooks, mechanics and stonemasonsWork & jobs · AI and jobs · arxiv.org ↗
- In the GPT-4 system card, OpenAI describes how during ARC's testing the model asked a TaskRabbit worker to solve a CAPTCHA and, when asked whether it was a robot, claimed a vision impairmentCybersecurity · Software development · Death of CAPTCHA · cdn.openai.com ↗
- GPT-4 is releasedScience · Business & markets · Large language model · openai.com ↗
- ChatGPT becomes the fastest-growing consumer app: ~100 million users in two months (UBS estimate)Business & markets · Chatbots · finance.yahoo.com ↗
- ChatGPT launchesEducation · Work & jobs · Media & information · Chatbots · openai.com ↗
- Programmers file a US class action against GitHub, Microsoft and OpenAI over Copilot — the first case challenging the training and output of AI systemsSoftware development · Laws & regulation · AI pair programmer · githubcopilotlitigation.com ↗
- Whisper: open speech recognitionMedia & information · Software development · Speech recognition · en.wikipedia.org ↗
- OpenAI describes InstructGPT: the model is fine-tuned on human ratings (RLHF), and people prefer the answers of its 1.3-billion-parameter version to those of the 175-billion-parameter GPT-3Science · Learning from human feedback (RLHF) · arxiv.org ↗
- GitHub Copilot: AI completes code right in the editorSoftware development · Work & jobs · AI pair programmer · github.blog ↗
- DALL·E: images from a text descriptionCulture & arts · Generative AI · en.wikipedia.org ↗
- In GPT-3's training data Wikipedia is only 3% of the mix (3 billion tokens), yet the model saw it 3.4 times, versus less than once (0.44) for the bulk Common Crawl web dataMedia & information · Science · Wikipedia after AI · arxiv.org ↗
- GPT-3: 175 billion parameters, solves tasks from a few examplesScience · Large language model · arxiv.org ↗
- Jared Kaplan and co-authors at OpenAI describe scaling laws: a language model's loss falls as a power law as the number of parameters, the amount of data and training compute growScience · Large language model · arxiv.org ↗
- Microsoft invests $1 billion in OpenAIBusiness & markets · AI labs · globalventuring.com ↗
- GPT-2: OpenAI initially withholds the full model over the risk of misuseScience · Media & information · Large language model · openai.com ↗
- OpenAI GPT: a transformer pre-trained on textScience · Large language model · openai.com ↗
- OpenAI is foundedScience · Business & markets · AI labs · openai.com ↗