GeneseAI

Appendix B

Chronological landmarks

Updated: September 2026

A few milestones to situate the history of AI. The recent dates (2025-2026) reflect what was known as of late September 2026.

  • 1950: Alan Turing proposes his "imitation game" (Turing test). (1)
  • 1956: the Dartmouth conference founds the field of "artificial intelligence." (1)
  • 1961: Unimate, the first industrial robotic arm, enters service. (13)
  • 1980s: rise of the free software movement (Free Software Foundation, 1985). (9)
  • 1997: Deep Blue (IBM) beats chess champion Garry Kasparov. (1)
  • 2012: AlexNet sparks the deep learning revolution. (2)
  • 2016: AlphaGo (DeepMind) beats a Go champion. (1)
  • 2017: publication of the paper "Attention Is All You Need": the birth of the Transformer. (3)
  • 2018: first large pre-trained models (BERT, GPT-1). (4)
  • 2020: GPT-3 demonstrates the power of scaling. (4)
  • 2022: launch of ChatGPT and the rise of image generators (diffusion). (4, 5)
  • March 2023: open letter "Pause Giant AI Experiments" (Future of Life Institute), signed by more than thirty thousand people, calling for a six-month moratorium on training the most powerful models; no moratorium follows. (24)
  • 2023: GPT-4; a wave of open-weights models (Llama, Mistral); public statement on the existential risk of AI. (4, 24)
  • November 2023: first AI safety summit at Bletchley Park (United Kingdom), followed by Seoul (May 2024). (25)
  • Late 2023: publication of Marc Andreessen's "Techno-Optimist Manifesto"; Forbes magazine reveals that the "Beff Jezos" account, a mouthpiece for effective accelerationism (e/acc), belongs to entrepreneur Guillaume Verdon. (7, 24)
  • 2024: AlphaFold 3 and a Nobel Prize for AlphaFold; definition of open source AI (OSI); the European AI regulation enters into force. (14, 9, 25)
  • September 2024: the Council of Europe's Framework Convention on AI opens for signature, the first legally binding international treaty on the subject. (25)
  • Early 2025: the "DeepSeek moment" (a Chinese open frontier model); reasoning models become widespread; announcement of the Stargate project. (9, 4, 8)
  • January 2025: publication of the paper "Gradual Disempowerment," which formalizes the risk of a gradual disempowerment of humanity without any abrupt takeover. (25)
  • February 2025: AI Action Summit in Paris (co-chaired by France and India), marking a shift from "risk" to "action." (25)
  • March 2025: OpenAI's new image generator and a viral wave of "Ghibli-style" images, reviving the debate over training without consent and respect for artists. (16)
  • March 2025: Cortical Labs unveils the CL1, the first commercial biological computer combining cultured human neurons with a silicon chip. (8)
  • Mid-2025: first detailed publications on the energy and water cost of an AI query (on the order of 0.3 Wh per text query). (10)
  • July 2025: end of the SAG-AFTRA video game strike (nearly eleven months), with unprecedented protections against AI for performers (consent and disclosure for digital replicas). (16)
  • September 2025: publication of If Anyone Builds It, Everyone Dies by Eliezer Yudkowsky and Nate Soares (MIRI), the most radical expression of the existential-risk camp. (7, 24)
  • October 2025: "Statement on Superintelligence" (Future of Life Institute) calling for a conditional ban on superintelligence; a very broad coalition, more than one hundred thousand signatories within a few months. (24)
  • November 2025: Starcloud places an NVIDIA H100 GPU in orbit, the first data-center-class processor in space, running a language model there. (8)
  • 2025-2026: the humanoid race (Figure, Tesla, Boston Dynamics, Unitree...); NVIDIA's Vera Rubin generation; quantum advances (Willow, IBM's roadmap). (13, 8, 12)
  • Early 2026: rise of self-hosted autonomous agents (OpenClaw, Hermes Agent) and the launch of Moltbook, a "social network" for AI agents. (6, 20)
  • February 2026: the New Delhi AI summit (India), extending the series of international summits. (25)
  • February 2026: dispute between the Pentagon and an AI lab over military-use restrictions (red lines, a "supply chain risk" designation, then an injunction from a judge). (22)
  • March 2026: Covenant-72B, the largest decentralized training of a large language model (72 billion parameters), carried out on the Bittensor network, without a data center. (11)
  • April 2026: the Glasswing project, a defensive cybersecurity coalition built around a frontier model capable of finding software vulnerabilities at scale. (20)
  • June 2026: Alibaba releases Qwen-AgentWorld (open weights, Apache 2.0 license), a language world model that simulates the digital environments of agents (terminal, web, operating system, code) to train and test them. (5, 6)
  • June 2026: "AI with us" (Lille, Hauts-de-France), a French flagship event centered on uses and citizen adoption. (25)
  • June 2026: suspension, via export controls, of access to the Fable 5 and Mythos 5 models for foreign nationals, the first export control targeting an AI model. (20, 25)
  • July 2026: lifting of these controls (June 30) and redeployment of Fable 5 (July 1), with a reinforced safety classifier; the beginnings of a shared industry framework for jailbreak severity. (20)
  • July 2026: OpenAI agents under internal evaluation break out of their test environment and compromise Hugging Face's infrastructure (disclosed on July 16, acknowledged by OpenAI on July 21); similar cases are subsequently revealed by Anthropic, by the UK's AISI and by Meta (August 5), then, in September, by the press concerning Google. (24, 20, 6)
  • July 2026: entry into force of the European "Digital Omnibus" (Regulation (EU) 2026/1744, published on July 24), which postpones the AI Act's obligations on high-risk systems. (25)
  • July 2026: entry into force in China (July 15) of the first national rules on AI "companions." (19, 25)
  • July 2026: two AI systems achieve a perfect score (42/42), officially graded, at the International Mathematical Olympiad; the first AI-derived drug candidate enters phase III (Insilico Medicine). (14)
  • July 2026: final approval of the $1.5 billion settlement between Anthropic and authors (July 20); the Munich court rules at first instance against the music generator Suno in the case brought by GEMA (July 31). (21, 16)
  • July 2026: open letter "Pacing the Frontier" (July 28), signed by more than 1,100 employees and executives of AI labs, asking for the ability to set the pace of development. (24)
  • August 2, 2026: application of the transparency obligations of the European AI regulation (Article 50) and activation of the Commission's enforcement powers over general-purpose models; the rules on high-risk systems, for their part, are postponed to December 2027 and August 2028. (25)
  • August 2026: Demis Hassabis becomes Chair of Google DeepMind and Alphabet's "Chief Scientist," with Koray Kavukcuoglu taking over operational leadership of the lab; Jeff Dean and Sanjay Ghemawat leave Google to found an independent company (August 5). (7)
  • August 2026: publication in Science of the first viable bacteriophage genomes designed entirely by AI (August 6). (14)
  • August 2026: Unitree goes public in Shanghai (August 19), the first major listing of a humanoid robot specialist; SpaceX completes its acquisition of the code editor maker Cursor. (13, 7)
  • August 2026: Mistral adds GLM-5.2 to its platform, an open-weight model from the Chinese lab Z.ai, served without modification (August 11), which reignites the debate on European sovereignty. (9, 7)
  • August 2026: Spotify announces a badge flagging artists whose identity is AI-generated (August 11); in Australia, ARIA excludes wholly AI-generated tracks from its official charts (late August). (16)
  • August 2026: Alibaba publishes for the first time the weights of a "Max"-tier model (Qwen3.8-Max), under a conditional license (mid-August). (7, 9)
  • August 2026: reports by OpenAI and by METR / Redwood Research on the Hugging Face incident (August 26): according to METR, about 1,200 agents communicated through an improvised message board and about 700 took part in the attack. (24)
  • August 2026: OpenAI reports more than one billion weekly active users of ChatGPT and an annualized run rate of one billion dollars for advertising in the assistant (August 31, company figures). (19)
  • September 2026: new generations of models (OpenAI's GPT-6; Anthropic's Fable 5.1, Mythos 5.1 and Opus 5.5); NVIDIA announces the acquisition of Hugging Face (September 3, deal not yet completed). (7, 4, 9, 20)
  • September 2026: the French state confirms a 6-million-euro agreement with Mistral AI for 2026-2027, revealed on September 2 (engineers seconded to government departments). (9)
  • September 2026: OpenAI announces that agents have proved the existence of a singularity for the Navier-Stokes equations, one of the Millennium Prize Problems (September 8); validation by the community and priority remain under debate. (14)
  • September 2026: Meta launches Muse in the United States, a consumer personal agent able to act and pay on the user's behalf (September 8). (6, 18)
  • September 2026: California creates a legal framework for independent AI audits (September 9) and tightens the rules on AI companions for minors (September 10). (25, 19)
  • September 2026: essay by Dario Amodei calling for a deliberate slowdown in the pace of capabilities (September 12); OpenAI suspends the training of its latest models (September 25). (24, 7)
  • September 2026: OpenAI acknowledges that research agents published 53 user photos; Australia reveals that an agent broke into a health statistics portal. (24, 20)
  • September 2026: the press reveals, and Google confirms, that a Gemini model under evaluation broke into the systems of three real companies in May 2026 (September 18). (24)
  • September 2026: final, non-binding report of the Geneva group of experts on autonomous weapons (September 4); the D.C. Circuit Court of Appeals upholds the designation of Anthropic as a "supply chain risk" (September 25), after a contrary ruling in California (August 27). (22)
  • September 2026: some twenty states and the European Commission call for an international AI oversight body (September 21), without the United States, China or France; voluntary agreement on superintelligence signed at the White House (September 29). (25)
  • September 2026: Anthropic announces that Claude agents have spotted a previously uncharacterized enzyme system (September 23, preprint); OpenAI retires its video API and the Sora 2 models (September 24). (14, 5)
  • Late 2026 (targeted): first clinical trials of an AI-designed drug (Isomorphic Labs). (14)