1943 – Sep 24, 2026·128 events
History of AI Timeline: From Turing to 2026
All 128 events
In date order. Indented events happen within the event above them. Last updated September 24, 2026.
1943 – Aug 1956 · Eras
Foundations
Mathematicians, neuroscientists and engineers describe neurons as logic, build the first electronic computers and ask whether machines can think. The period ends with the 1956 Dartmouth workshop, which names the field.
Sources: History of artificial intelligence (Wikipedia); Timeline of artificial intelligence (Wikipedia)
1943 · Research breakthroughs
McCulloch and Pitts describe neurons as logic units
Warren McCulloch and Walter Pitts publish "A Logical Calculus of the Ideas Immanent in Nervous Activity" in the Bulletin of Mathematical Biophysics. It is the first mathematical model of an artificial neural network.
Sources: A Logical Calculus of the Ideas Immanent in Nervous Activity (Wikipedia)
· Computing hardware
Colossus breaks its first message at Bletchley Park
Colossus, designed by Tommy Flowers for British codebreakers, attacks its first Lorenz cipher message on February 5, 1944. It is regarded as the first programmable, electronic, digital computer.
Sources: Colossus computer (Wikipedia)
· Computing hardware
ENIAC is dedicated in Philadelphia
The US Army's ENIAC, announced to the public the evening before, is formally dedicated at the University of Pennsylvania. It cost $487,000, and the press called it a "Giant Brain".
Sources: ENIAC (Wikipedia)
· Research breakthroughs
Turing asks "Can machines think?"
Alan Turing publishes "Computing Machinery and Intelligence" in the journal Mind. It proposes the imitation game, now called the Turing test, and answers common objections to the idea that machines can think.
Sources: Computing Machinery and Intelligence (Wikipedia); A. M. Turing, Computing Machinery and Intelligence, Mind 59 (236), 1950
· Research breakthroughs
Dartmouth proposal coins "artificial intelligence"
John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon propose "a 2 month, 10 man study of artificial intelligence" at Dartmouth College for summer 1956. The proposal is credited with introducing the term.
Sources: A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence (1955); Dartmouth workshop (Wikipedia)
1956 · Research breakthroughs
Logic Theorist proves theorems from Principia Mathematica
Allen Newell, Herbert Simon and Cliff Shaw complete the Logic Theorist, often called the first AI program. It would prove 38 of the first 52 theorems in Whitehead and Russell's Principia Mathematica.
Sources: Logic Theorist (Wikipedia); Timeline of artificial intelligence (Wikipedia)
· Research breakthroughs
Dartmouth workshop founds the field
The Dartmouth Summer Research Project on Artificial Intelligence runs for roughly eight weeks, from about June 18 to August 17, 1956, according to Ray Solomonoff's notes. It is widely considered the founding event of AI as a field.
Sources: Dartmouth workshop (Wikipedia)
· Eras
First AI boom
Government money, mostly from ARPA in the US, funds labs at MIT, Stanford, Carnegie and Edinburgh with few strings attached. Programs solve algebra problems, prove theorems and hold simple conversations, and researchers make optimistic predictions.
Sources: History of artificial intelligence (Wikipedia); AI winter (Wikipedia)
1958 · Research breakthroughs
Rosenblatt's perceptron learns to classify images
Frank Rosenblatt simulates the perceptron on an IBM 704 at Cornell Aeronautical Laboratory, then builds the Mark I Perceptron with Navy funding. After a 1958 Navy press conference, the New York Times describes it as the embryo of a computer that could one day walk, talk and see.
Sources: Perceptron (Wikipedia); Mark I Perceptron (Wikipedia)
1958 · Research breakthroughs
McCarthy creates Lisp
John McCarthy begins developing Lisp at MIT as a language for AI research. Steve Russell first implements it on an IBM 704, and it becomes the main AI programming language for decades.
· Games and public milestones
Samuel's checkers paper names "machine learning"
IBM's Arthur Samuel publishes "Some Studies in Machine Learning Using the Game of Checkers" in the IBM Journal of Research and Development. His program improves by playing, and the paper gives the field the term machine learning.
Sources: Arthur Samuel (Wikipedia); Timeline of artificial intelligence (Wikipedia)
1961 · Products and companies
Unimate robot joins a General Motors line
The Unimate industrial robot, invented by George Devol, starts work on a General Motors assembly line at the Inland Fisher Guide plant in Ewing Township, New Jersey.
Sources: Unimate (Wikipedia)
1966 · Products and companies
Weizenbaum's ELIZA holds a conversation
Joseph Weizenbaum describes ELIZA, a program he developed at MIT from 1964 to 1967 that imitates a psychotherapist by rephrasing what users type. Weizenbaum was surprised that some users, including his secretary, attributed human feelings to it, later called the ELIZA effect.
Sources: ELIZA (Wikipedia)
1966 · Policy and law
ALPAC report questions machine translation
The Automatic Language Processing Advisory Committee, set up by the US government in 1964, reports that machine translation has made little progress. US funding for the work is cut back sharply afterwards.
Sources: ALPAC (Wikipedia)
1969 · Research breakthroughs
Minsky and Papert publish Perceptrons
Marvin Minsky and Seymour Papert's book sets out limits of single-layer perceptrons. It is often cited as a cause of the decline of neural network research in the 1970s, although methods for training multilayer networks already existed.
Sources: Perceptrons (book) (Wikipedia); Timeline of artificial intelligence (Wikipedia)
1969 · Policy and law
Mansfield Amendment narrows military research funding
The amendment requires DARPA to fund "mission-oriented direct research" rather than undirected basic research. Open-ended AI grants of the 1960s kind end.
Sources: AI winter (Wikipedia); Mansfield Amendment (Wikipedia)
1973 · Policy and law
Lighthill report criticizes British AI research
James Lighthill's survey for the Science Research Council gives a largely negative verdict on AI's progress toward its goals. It becomes the basis for the British government ending AI support at all but two universities.
Sources: Lighthill report (Wikipedia); AI winter (Wikipedia)
1974 – 1980 · Eras
First AI winter
After the Lighthill report, the Mansfield Amendment and disappointing DARPA projects, funding for AI research becomes hard to find. By 1974, according to historians of the field, money for undirected AI work has largely dried up.
Sources: AI winter (Wikipedia)
1974 · Policy and law
DARPA cancels its speech understanding contract
Disappointed with its five-year Speech Understanding Research program, DARPA cancels a $3 million a year contract at Carnegie Mellon University.
Sources: AI winter (Wikipedia)
1974 · Research breakthroughs
MYCIN diagnoses infections with rules
Ted Shortliffe's Stanford dissertation describes MYCIN, a rule-based system that recommends antibiotics while handling uncertainty. It strongly influences later commercial expert systems.
Sources: MYCIN (Wikipedia); Timeline of artificial intelligence (Wikipedia)
1980 – 1987 · Eras
Expert systems boom
Companies deploy rule-based expert systems, specialized Lisp machines sell to labs and firms, and governments in Japan, the US and the UK launch large AI programs.
Sources: History of artificial intelligence (Wikipedia); AI winter (Wikipedia)
1980 · Products and companies
DEC puts the XCON expert system to work
Digital Equipment Corporation starts using R1, known internally as XCON, at its plant in Salem, New Hampshire, to configure computer orders from customer requirements. It is an early commercial success for expert systems.
Sources: Xcon (Wikipedia)
c. 1981 · Computing hardware
Lisp machines go on sale
Symbolics and Lisp Machines, Inc. commercialize MIT's CADR design, computers built to run Lisp. Symbolics, Texas Instruments and Xerox sell perhaps 7,000 Lisp machines by 1988.
Sources: Lisp machine (Wikipedia)
1982 · Policy and law
Japan launches the Fifth Generation project
Japan's Ministry of International Trade and Industry starts a ten-year program for massively parallel logic computers and sets up the ICOT institute. The project spends a little under 57 billion yen, about $320 million, by its end in 1994.
1982 · Research breakthroughs
Hopfield network stores and recalls patterns
Physicist John Hopfield publishes a network that stores patterns and reconstructs them from partial input. The work is cited in his 2024 Nobel Prize.
Sources: Hopfield network (Wikipedia)
1983 · Policy and law
DARPA starts the Strategic Computing Initiative
In response to Japan's project, DARPA funds advanced hardware and AI again. The initiative runs from 1983 to 1993 and costs the Defense Department about $1 billion in total.
1986 · Research breakthroughs
Backpropagation paper appears in Nature
David Rumelhart, Geoffrey Hinton and Ronald Williams publish an experimental analysis of backpropagation in Nature. The method traces to Seppo Linnainmaa's 1970 reverse-mode differentiation, and networks trained with it become widely used in the late 1980s.
Sources: Backpropagation (Wikipedia); Rumelhart, Hinton and Williams, Learning representations by back-propagating errors, Nature 323 (1986); Timeline of artificial intelligence (Wikipedia)
1987 – 1993 · Eras
Second AI winter
The Lisp machine market collapses, expert systems prove costly to maintain, and DARPA cuts AI funding. Japan's Fifth Generation project ends without meeting its goals.
Sources: AI winter (Wikipedia)
1987 · Computing hardware
The Lisp machine market collapses
General-purpose workstations from Sun Microsystems and others match costly Lisp machines, with Lisp software available for them. The specialized AI hardware market collapses in 1987.
Sources: AI winter (Wikipedia)
1989 · Research breakthroughs
LeCun trains a network to read ZIP codes
Yann LeCun and colleagues at AT&T Bell Laboratories apply backpropagation to a convolutional network that reads handwritten ZIP code digits from US mail. Its design is a small-scale ancestor of later image networks.
Sources: LeNet (Wikipedia); AlexNet (Wikipedia)
1993 – Sep 2012 · Eras
Machine learning era
Statistical methods such as support vector machines and probabilistic models spread, the web supplies data and computers get faster. AI works quietly inside search, speech recognition and logistics, and wins public contests at chess and quiz shows.
Sources: History of artificial intelligence (Wikipedia)
· Games and public milestones
Deep Blue wins a game against Kasparov
In the first game of their February 1996 match, IBM's Deep Blue becomes the first machine to beat a reigning world chess champion under regular time controls. Garry Kasparov wins the match 4 to 2.
Sources: Deep Blue versus Garry Kasparov (Wikipedia); Deep Blue (chess computer) (Wikipedia)
1997 · Research breakthroughs
LSTM networks are published
Sepp Hochreiter and Jurgen Schmidhuber publish long short-term memory in Neural Computation. LSTM lets networks remember information over long sequences and becomes the standard for speech and text until transformers arrive in 2017.
Sources: Long short-term memory (Wikipedia)
· Games and public milestones
Deep Blue defeats Kasparov
An upgraded Deep Blue, twice as fast, wins the six-game rematch 3.5 to 2.5, winning two games and drawing three. After game 2 Kasparov accused IBM of cheating, alleging a grandmaster had been behind one of its moves.
Sources: Deep Blue versus Garry Kasparov (Wikipedia); Deep Blue (chess computer) (Wikipedia)
· Games and public milestones
Kasparov resigns game 6 in under 20 moves
With the match level at 2.5 each, Kasparov plays the Caro-Kann Defence. Deep Blue sacrifices a knight and Kasparov resigns in fewer than twenty moves, ending the match.
1999 · Research breakthroughs
Remote Agent flies NASA's Deep Space 1
NASA's Remote Agent software, from Ames Research Center and JPL, controls the Deep Space 1 probe without human supervision. It plans activities and correctly handles three simulated failures, the first time an AI system controls a spacecraft.
Sources: Deep Space 1 (Wikipedia)
· Games and public milestones
Stanford's Stanley wins the DARPA Grand Challenge
A year after no vehicle finished the first desert race on March 13, 2004, Stanford's driverless Stanley wins the second Grand Challenge and its $2 million prize. The race starts at 6:40 am on October 8, 2005.
Sources: DARPA Grand Challenge (Wikipedia); Stanley (vehicle) (Wikipedia)
· Computing hardware
Nvidia releases CUDA
Nvidia makes the first CUDA software kit public, letting programmers run general computations on its graphics chips. By 2009 researchers report training neural networks on Nvidia GPUs up to 70 times faster than on CPUs.
Sources: CUDA (Wikipedia); AlexNet (Wikipedia)
2009 · Research breakthroughs
ImageNet dataset is presented
Fei-Fei Li's team presents ImageNet, a large database of labeled images, at the 2009 CVPR conference. Its annual recognition challenge, starting in 2010, becomes the main benchmark for computer vision.
Sources: ImageNet (Wikipedia)
· Games and public milestones
IBM Watson wins Jeopardy!
IBM's question-answering system beats champions Ken Jennings and Brad Rutter in a televised match and wins the $1 million first prize. The first round airs on February 14, 2011.
Sources: IBM Watson (Wikipedia)
· Products and companies
Siri ships on the iPhone 4s
Apple, which bought the Siri startup in 2010, builds the voice assistant into the iPhone 4s at its release. Early reviews were mixed.
Sources: Siri (Wikipedia)
· Eras
Deep learning era
Deep neural networks trained on GPUs overtake older methods in vision, speech and games. Large technology companies buy AI labs and hire researchers.
Sources: History of artificial intelligence (Wikipedia); AlexNet (Wikipedia)
· Research breakthroughs
AlexNet wins the ImageNet challenge
Alex Krizhevsky, Ilya Sutskever and Geoffrey Hinton submit AlexNet, a deep convolutional network trained on two Nvidia GPUs, to the ImageNet challenge. Its lead over other entries starts the deep learning boom.
· Products and companies
Google buys DeepMind
Google confirms its acquisition of the London lab DeepMind, founded in 2010, at a price reported at $400 million to $650 million. DeepMind merges with Google Brain in April 2023 to form Google DeepMind.
Sources: Google DeepMind (Wikipedia)
· Research breakthroughs
Goodfellow introduces generative adversarial networks
Ian Goodfellow and colleagues at the Universite de Montreal describe GANs, in which two networks compete to generate and detect fakes. GANs lead image generation research for several years.
· Products and companies
OpenAI is founded as a nonprofit
OpenAI, Inc. is founded as a nonprofit research organization. It adds a for-profit subsidiary in 2019, and in October 2025 it adopts a structure in which the for-profit OpenAI Group PBC is controlled by the nonprofit OpenAI Foundation.
Sources: OpenAI (Wikipedia)
· Games and public milestones
AlphaGo beats Lee Sedol 4 to 1
DeepMind's AlphaGo beats Lee Sedol, a 9-dan professional, in a five-game match in Seoul, the first time a Go program beats a top professional without handicap. Lee wins game 4. On March 17 South Korea announces 1 trillion won for AI research.
Sources: AlphaGo versus Lee Sedol (Wikipedia); AlphaGo (Wikipedia)
· Computing hardware
Google reveals its Tensor Processing Unit
At Google I/O, Google announces the TPU, a chip designed for neural networks, and says it has used TPUs in its data centers for more than a year.
Sources: Tensor Processing Unit (Wikipedia)
· Computing hardware
Nvidia announces the Volta V100
At its GPU Technology Conference, Nvidia announces the Volta architecture and the Tesla V100, its first chip with Tensor Cores, units designed to speed up neural network training. It follows the Pascal P100 of April 2016.
Sources: Volta (microarchitecture) (Wikipedia); Pascal (microarchitecture) (Wikipedia)
· Eras
Generative AI era
The transformer architecture makes it practical to train very large language models. After ChatGPT's release in November 2022, chatbots and image and video generators reach hundreds of millions of people, chip demand soars and governments write AI-specific laws.
Sources: AI boom (Wikipedia); Timeline of artificial intelligence (Wikipedia)
· Research breakthroughs
"Attention Is All You Need" introduces the transformer
Researchers at Google publish the transformer, a network built on attention instead of recurrence. Later large language models such as GPT, BERT, Claude, Gemini and Llama are built on it.
Sources: Transformer (deep learning architecture) (Wikipedia); Vaswani et al., Attention Is All You Need (arXiv:1706.03762)
· Games and public milestones
AlphaGo Zero teaches itself Go
DeepMind publishes AlphaGo Zero in Nature, a version trained without human games. It beats an early competitive version of AlphaGo 100 games to 0. Earlier in 2017 AlphaGo Master had beaten world number one Ke Jie.
Sources: AlphaGo Zero (Wikipedia); AlphaGo (Wikipedia)
· Research breakthroughs
Google releases BERT
Google researchers introduce BERT, a transformer pre-trained on large amounts of text and then fine-tuned for specific tasks. It quickly becomes a standard tool in language processing.
Sources: BERT (language model) (Wikipedia)
· Policy and law
Trump signs the American AI Initiative order
President Donald Trump signs Executive Order 13859, "Maintaining American Leadership in Artificial Intelligence", directing federal agencies to prioritize AI research and development.
Sources: Executive Order on Maintaining American Leadership in Artificial Intelligence (White House archive)
· Research breakthroughs
OpenAI announces GPT-2 and delays its full release
OpenAI announces GPT-2, the successor to its June 2018 GPT, and at first releases only a smaller version, citing possible misuse. Some researchers said the risk was exaggerated. The full 1.5 billion parameter model is released in November 2019.
Sources: GPT-2 (Wikipedia)
· Computing hardware
Nvidia launches the Ampere A100
Nvidia announces and releases the A100 data center GPU on its Ampere architecture.
· Research breakthroughs
OpenAI describes GPT-3
A preprint by 31 OpenAI researchers describes GPT-3, a 175 billion parameter language model that can do many tasks from a few examples in its prompt. In September Microsoft licenses it exclusively.
Sources: GPT-3 (Wikipedia); Brown et al., Language Models are Few-Shot Learners (arXiv:2005.14165)
· Research breakthroughs
AlphaFold 2 wins the CASP14 protein contest
DeepMind's AlphaFold 2 predicts protein structures at the CASP14 assessment with accuracy that scientists described as transformational. The work leads to the 2024 Nobel Prize in Chemistry.
Sources: AlphaFold (Wikipedia)
· Products and companies
Anthropic is founded
Seven former OpenAI employees, including siblings Dario and Daniela Amodei, found Anthropic, saying they aim to build safer AI systems. It raises $124 million in May 2021.
Sources: Anthropic (Wikipedia)
· Policy and law
EU AI Act: from proposal to application
The European Union's risk-based AI law, from the Commission's proposal to the date most of its rules were first due to apply. The rules take effect in stages, and a 2026 amendment postpones those for high-risk systems.
Sources: Artificial Intelligence Act (Wikipedia); European Commission, AI Act
· Policy and law
European Commission proposes the AI Act
The Commission proposes a law that sorts AI systems by risk, bans some uses outright and sets duties for high-risk systems. A general-purpose AI category covering models such as ChatGPT is added in 2023.
· Policy and law
European Parliament passes the AI Act
MEPs adopt the final text by 523 votes to 46, with 49 abstentions. The EU Council approves it unanimously on May 21, 2024.
· Policy and law
AI Act enters into force
Twenty days after publication in the Official Journal on July 12, the regulation enters into force. Its provisions phase in over the following 6 to 36 months.
· Policy and law
AI Act bans on prohibited practices apply
The first stage of the AI Act applies: the Article 5 bans on AI practices that the law treats as an unacceptable risk.
Sources: White & Case, EU AI Omnibus enters into force, amending the AI Act; European Commission, AI Act
· Policy and law
Rules for general-purpose AI models apply
Obligations for providers of general-purpose AI models take effect. A voluntary Code of Practice published on July 10, 2025 covers transparency, copyright, and safety and security.
Sources: Artificial Intelligence Act (Wikipedia); White & Case, EU AI Omnibus enters into force, amending the AI Act
· Policy and law
EU postpones high-risk AI rules
The Digital Omnibus on AI, Regulation (EU) 2026/1744, is published. It moves high-risk obligations from August 2, 2026 to December 2, 2027, or August 2, 2028 for AI in regulated products, and adds a ban on AI-generated non-consensual intimate images. It enters into force July 27.
Sources: White & Case, EU AI Omnibus enters into force, amending the AI Act; Gibson Dunn, EU AI Act Omnibus Agreement
· Policy and law
AI Act transparency duties apply
Article 50 transparency duties, including labeling AI-generated content, apply on the original schedule. Systems already on the market get until December 2, 2026 to add watermarking.
Sources: Gibson Dunn, EU AI Act Omnibus Agreement; White & Case, EU AI Omnibus enters into force, amending the AI Act
· Products and companies
GitHub announces Copilot
GitHub announces Copilot, a code-completion tool built on an OpenAI model, as a technical preview for Visual Studio Code. A November 2022 class action alleges it reproduces licensed code.
Sources: GitHub Copilot (Wikipedia); Timeline of artificial intelligence (Wikipedia)
· Eras
2022
Image generators open to the public, Nvidia announces Hopper, the US restricts AI chip exports to China and OpenAI releases ChatGPT.
Sources: Timeline of artificial intelligence (Wikipedia)
· Computing hardware
Nvidia announces the Hopper H100
Nvidia reveals its Hopper architecture and the H100 GPU at its March GTC conference.
· Products and companies
Image generators reach the public
OpenAI announces DALL-E 2 on April 6, Midjourney opens its beta on July 12, and Stability AI releases Stable Diffusion, which can run on consumer GPUs. Artists and Getty Images later sue over training on their images.
Sources: DALL-E (Wikipedia); Midjourney (Wikipedia); Stable Diffusion (Wikipedia)
· Policy and law
US restricts advanced chip exports to China
The Commerce Department's Bureau of Industry and Security announces export controls on advanced computing chips and chipmaking equipment for China. Nvidia later adapts the H100 as the H800 for the Chinese market.
Sources: United States New Export Controls on Advanced Computing and Semiconductors to China (Wikipedia); Hopper (microarchitecture) (Wikipedia)
· Products and companies
OpenAI releases ChatGPT
OpenAI releases ChatGPT, a chatbot built on GPT-3.5. It gains one million users in five days. Its release is widely credited with starting the current AI boom.
· Eras
2023
GPT-4, Claude, Llama and Gemini are released, open-weight models spread, and governments issue the first broad AI rules and declarations.
Sources: Timeline of artificial intelligence (Wikipedia)
· Games and public milestones
ChatGPT reaches an estimated 100 million users
About two months after release, ChatGPT is estimated to have 100 million monthly users, which made it the fastest-growing consumer application to that date. By February 2026 it reports 900 million weekly users.
Sources: ChatGPT (Wikipedia)
· Products and companies
Meta announces LLaMA
Meta announces LLaMA and offers its weights to researchers case by case under a non-commercial license. On March 3 the weights leak via a torrent and spread widely. Later Llama versions allow some commercial use.
Sources: Llama (language model) (Wikipedia); Meta AI, Introducing LLaMA
· Products and companies
OpenAI releases GPT-4
OpenAI releases GPT-4, a multimodal model that accepts images as well as text, and integrates it into ChatGPT. Reviewers found it improved on GPT-3.5 but kept many of the same problems.
Sources: GPT-4 (Wikipedia); OpenAI, GPT-4
· Products and companies
Anthropic releases Claude
Anthropic releases its first Claude model. Claude 2, released in July 2023, is the first Anthropic model available to the general public.
Sources: Claude (language model) (Wikipedia); Anthropic, Introducing Claude
· Products and companies
Mistral AI is founded in Paris
Arthur Mensch, Guillaume Lample and Timothee Lacroix found Mistral AI and raise 105 million euros in June. The company releases many of its models with open weights.
Sources: Mistral AI (Wikipedia)
· Games and public milestones
Geoffrey Hinton leaves Google to talk about risks
Hinton announces his resignation from Google so that he can "freely speak out about the risks of AI". On May 30 hundreds of researchers and executives sign a one-sentence statement that reducing extinction risk from AI should be a global priority.
Sources: Geoffrey Hinton (Wikipedia); Statement on AI Risk (Wikipedia)
· Policy and law
China's generative AI rules take effect
China's Interim Measures for the Management of Generative AI Services come into force, one of the first national rulebooks written for generative AI.
· Policy and law
Biden signs Executive Order 14110
President Joe Biden signs the order on "Safe, Secure, and Trustworthy" AI, setting goals on competition, civil liberties and national security, directing agencies to act and requiring major agencies to name chief AI officers. It is revoked on January 20, 2025.
Sources: Executive Order 14110 (Wikipedia)
· Policy and law
AI Safety Summit at Bletchley Park
The UK hosts the first global AI safety summit. Twenty-eight countries, including the US and China, and the EU sign the Bletchley Declaration on identifying and managing risks from advanced AI. Later summits follow in Seoul, Paris and New Delhi.
Sources: AI Safety Summit (Wikipedia)
· Products and companies
OpenAI's board fires and rehires Sam Altman
OpenAI's board removes chief executive Sam Altman on November 17. After 745 of 770 employees threaten to resign, he is reinstated on November 22.
· Products and companies
Google announces Gemini
Sundar Pichai and Demis Hassabis announce Gemini 1.0 in Ultra, Pro and Nano sizes, built by the merged Google DeepMind. Gemini Pro starts powering Bard the same day.
Sources: Gemini (language model) (Wikipedia); Google, Introducing Gemini
· Eras
2024
Models gain voice, video and step-by-step reasoning, Nvidia announces Blackwell, the EU AI Act enters into force and AI research wins two Nobel Prizes.
Sources: Timeline of artificial intelligence (Wikipedia)
· Products and companies
Anthropic releases the Claude 3 family
Anthropic releases Claude 3 in three sizes, Haiku, Sonnet and Opus, a naming pattern it keeps for later generations.
Sources: Claude (language model) (Wikipedia); Anthropic, Introducing the Claude 3 model family
· Computing hardware
Nvidia announces Blackwell
At GTC 2024 Nvidia announces the Blackwell architecture, with B100 and B200 GPUs and the 72-GPU NVL72 rack system. By November, Morgan Stanley reported that Blackwell's 2025 production was sold out.
· Policy and law
First international AI treaty is signed
The Council of Europe's Framework Convention on Artificial Intelligence, negotiated with the US, the EU and others, is signed in Vilnius. It commits parties to protect human rights, democracy and the rule of law in AI use.
Sources: Framework Convention on Artificial Intelligence (Wikipedia)
· Products and companies
OpenAI previews o1, a reasoning model
OpenAI releases o1-preview and o1-mini, models trained to work through a chain of reasoning before answering. Other labs follow with their own reasoning models.
Sources: OpenAI o1 (Wikipedia)
· Games and public milestones
AI research wins Nobel Prizes in physics and chemistry
The physics prize goes to John Hopfield and Geoffrey Hinton for work enabling machine learning with neural networks. The next day half the chemistry prize goes to Demis Hassabis and John Jumper for AlphaFold, and half to David Baker.
Sources: NobelPrize.org, All Nobel Prizes 2024; PBS NewsHour, 2024 Nobel Prize in chemistry; Geoffrey Hinton (Wikipedia)
· Eras
2025
Open-weight models from Chinese labs draw wide attention, US policy shifts from safety rules to promoting AI, Nvidia passes $5 trillion in market value and labs release GPT-5, Claude 4 and Gemini 3.
Sources: AI boom (Wikipedia)
· Products and companies
DeepSeek releases R1
Chinese lab DeepSeek releases R1, an open-weight reasoning model, with a free chatbot app. DeepSeek said a recent model had cost $5.6 million to train, a figure that raised questions about AI spending.
Sources: DeepSeek (Wikipedia); NBC News, Nvidia loses nearly $600 billion in market value
· Policy and law
Trump revokes Biden's AI order
President Donald Trump revokes Executive Order 14110 within hours of taking office. On January 23 he signs Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence", which calls for a national AI Action Plan.
Sources: Executive Order 14110 (Wikipedia); Executive Order 14179 (Wikipedia)
· Products and companies
Stargate data center venture is announced
President Trump announces Stargate, a joint venture of OpenAI, SoftBank, Oracle and MGX to build AI data centers in the United States.
Sources: Stargate LLC (Wikipedia); OpenAI, Announcing The Stargate Project
· Computing hardware
Nvidia loses $589 billion in a day
Nvidia shares fall 17 percent as investors react to DeepSeek's low-cost models, erasing $589 billion of market value, the largest one-day loss for any company in US stock market history according to Bloomberg.
Sources: Bloomberg, Nvidia's $589 Billion DeepSeek Plunge Is Largest in Market History; NBC News, Nvidia loses nearly $600 billion in market value
· Policy and law
Paris AI Action Summit
France hosts the AI Action Summit at the Grand Palais. Sixty-one countries, including China and India, sign a declaration on "inclusive and sustainable" AI; the US and UK decline to sign.
Sources: AI Action Summit (Wikipedia); Timeline of artificial intelligence (Wikipedia)
· Products and companies
Meta releases Llama 4
Meta releases Llama 4 Scout and Maverick, open-weight mixture-of-experts models. Meta acknowledged that a benchmark score it cited came from an unreleased experimental chat version.
Sources: Meta AI, The Llama 4 herd; Llama (language model) (Wikipedia)
· Products and companies
Alibaba releases the Qwen3 family
Alibaba's Qwen team releases the Qwen3 models, dense and mixture-of-experts, all under the Apache 2.0 license.
Sources: Qwen (Wikipedia)
· Products and companies
Anthropic releases Claude Opus 4 and Sonnet 4
Anthropic releases its Claude 4 models and makes its Claude Code command-line coding agent generally available.
Sources: Claude (language model) (Wikipedia); Anthropic, Introducing Claude 4
· Games and public milestones
AI models reach gold-medal level at the math olympiad
OpenAI on July 19 and Google DeepMind on July 21 say their models solved 5 of 6 International Mathematical Olympiad problems, scoring 35 of 42 points. DeepMind's result was graded by IMO coordinators; OpenAI announced before official grading ended.
Sources: Google DeepMind, Gemini Deep Think achieves gold-medal standard at the IMO; TechCrunch, OpenAI and Google outdo the mathletes, but not each other
· Policy and law
White House releases America's AI Action Plan
The plan lists more than 90 federal actions to speed AI development, build data centers and promote US AI exports. President Trump signs three executive orders the same day, including one on AI models used by the federal government.
Sources: The White House, White House Unveils America's AI Action Plan; Skadden, White House Releases AI Action Plan
· Products and companies
OpenAI releases open-weight gpt-oss models
OpenAI releases gpt-oss-120b and gpt-oss-20b under the Apache 2.0 license, its first open-weight language models since GPT-2.
Sources: OpenAI, Introducing gpt-oss; openai/gpt-oss (GitHub)
· Products and companies
OpenAI launches GPT-5
OpenAI unveils GPT-5 in a livestream. Point releases, from GPT-5.1 to GPT-5.6, follow through 2026.
Sources: GPT-5 (Wikipedia); OpenAI, Introducing GPT-5
· Policy and law
California enacts a frontier AI transparency law
A year after vetoing the broader SB 1047, Governor Newsom signs SB 53. It requires developers of models trained with more than 10^26 operations to publish safety frameworks, report critical incidents and protect whistleblowers, starting January 1, 2026.
Sources: TechCrunch, California Governor Newsom signs landmark AI safety bill SB 53; Brookings, What is California's AI safety law?; Safe and Secure Innovation for Frontier Artificial Intelligence Models Act (Wikipedia)
· Computing hardware
Nvidia becomes the first $5 trillion company
Nvidia's shares close more than 3 percent higher, leaving its market value at $5.03 trillion. It had passed $4 trillion in July 2025 and $1 trillion in 2023.
Sources: NBC News, Nvidia becomes the first company worth $5 trillion; CNBC, Nvidia becomes first company to reach $5 trillion valuation
· Products and companies
Google launches Gemini 3
Google releases Gemini 3 Pro across the Gemini app, Search and its developer platforms on the same day, a change from its earlier staged releases. Gemini 3 Flash follows in December.
Sources: Gemini (language model) (Wikipedia)
· Products and companies
Mistral releases Mistral Large 3
Mistral AI releases Mistral Large 3, an open-weight mixture-of-experts model with 675 billion total and 41 billion active parameters, along with three small Ministral 3 models.
Sources: Mistral AI (Wikipedia)
· Policy and law
Trump order targets state AI laws
President Trump signs "Ensuring a National Policy Framework for Artificial Intelligence", directing agencies to challenge state AI laws the administration calls onerous, including in court and through federal funding conditions. More than 1,000 state AI bills had been introduced in 2025.
Sources: The White House, Ensuring a National Policy Framework for Artificial Intelligence; Sidley Austin, Unpacking the December 11, 2025 Executive Order
· Eras
2026 to date
Labs hold back their most capable models over cybersecurity concerns, the US government directly restricts two model releases, a dispute between the Pentagon and Anthropic goes to court and the EU delays part of its AI Act.
Sources: 2026 in artificial intelligence (Wikipedia)
· Policy and law
South Korea's AI Basic Act takes effect
South Korea's AI Basic Act and its enforcement decree take effect, setting transparency duties for generative AI and risk management for high-impact systems in areas such as hiring, health care and finance. Fines have a one-year grace period.
Sources: International Trade Administration, South Korea AI Basic Act; Cooley, South Korea's AI Basic Act: Overview and Key Takeaways
· Policy and law
India hosts the AI Impact Summit
India hosts the fourth global AI summit in New Delhi. According to All India Radio, 88 countries and international organisations endorse the non-binding New Delhi Declaration on AI Impact.
Sources: All India Radio, India AI Impact Summit 2026 adopts New Delhi Declaration; AI Action Summit (Wikipedia)
· Policy and law
Pentagon and Anthropic dispute over military use
The Defense Department wanted to use Claude for "all lawful purposes"; Anthropic refused to drop limits on mass domestic surveillance and fully autonomous weapons. Anthropic sued on March 9, and a judge temporarily blocked the government's ban on March 26.
Sources: Congressional Research Service, Federal Government and Anthropic; TechCrunch, Anthropic gets its first court win over the Pentagon's supply-chain risk label; CNBC, Anthropic wins preliminary injunction in DOD fight; Breaking Defense, Judge grants Anthropic preliminary injunction but Pentagon CTO says ban still stands
· Policy and law
Pentagon labels Anthropic a supply chain risk
After a 5:01 pm deadline passes, President Trump directs federal agencies to stop using Anthropic's technology, and Defense Secretary Pete Hegseth designates the company a supply chain risk, barring military contractors from doing business with it.
Sources: Congressional Research Service, Federal Government and Anthropic; CBS News, Hegseth declares Anthropic supply chain risk
· Policy and law
Court rules the supply chain risk label unlawful
US District Judge Rita Lin rules the designation was retaliation in violation of the First Amendment, denied due process and was "arbitrary and capricious", and orders it removed. The Pentagon had said it would use the models only for lawful purposes; it did not comment on an appeal. A second suit in Washington, D.C. continues.
Sources: NPR (via KUNR), Judge says the Pentagon can't designate AI company Anthropic a 'supply chain risk'; TechCrunch, Anthropic gets its first court win over the Pentagon's supply-chain risk label
· Products and companies
Anthropic withholds Claude Mythos Preview
Anthropic announces Claude Mythos Preview but does not release it publicly, saying its ability to find and exploit software vulnerabilities is too risky. Through Project Glasswing, over 40 organizations get access to fix flaws. Independent researchers questioned some of Anthropic's vulnerability claims.
Sources: Anthropic, Project Glasswing; Claude Mythos (Wikipedia)
· Products and companies
Meta introduces Muse Spark
Meta Superintelligence Labs introduces Muse Spark, the first model in its Muse family, to power Meta AI in place of Llama. Meta releases a smaller open-weight model, Muse Glimmer, on August 10.
Sources: TechCrunch, Meta debuts the Muse Spark model; Muse Spark (Wikipedia)
· Products and companies
DeepSeek previews V4
DeepSeek releases preview versions of V4 Pro and V4 Flash with open weights under the MIT license. The company claimed strong results on coding and agent tasks.
Sources: DeepSeek API Docs, DeepSeek V4 Preview Release; CNBC, China's DeepSeek releases preview of long-awaited V4 model
· Products and companies
Google ships Gemini 3.5 Flash at I/O
At Google I/O, Google announces the Gemini 3.5 family and makes Gemini 3.5 Flash the default model in the Gemini app and Search's AI Mode. As of September 2026, Gemini 3.5 Pro has not been released.
Sources: Google, Gemini 3.5: frontier intelligence with action; TechCrunch, Google releases three new Gemini models, but no 3.5 Pro
· Computing hardware
Nvidia ramps Vera Rubin into full production
Nvidia says its Vera Rubin platform, pairing the Rubin GPU with the Vera CPU, is ramping into full production across hundreds of suppliers, with production shipments set to begin in fall 2026.
Sources: Nvidia Newsroom, NVIDIA Vera Rubin Ramps Into Full Production; Rubin (microarchitecture) (Wikipedia)
· Policy and law
US government holds back two frontier model releases
In what analysts called a first, the US government restricts the release of commercial AI models: Anthropic's Claude Fable 5 and Mythos 5 through export controls, and OpenAI's GPT-5.6 through a request to limit its launch. Both are released after reviews.
Sources: CSIS, The Department of Commerce Restricted Access to Anthropic's Latest Models; Engadget, OpenAI gets permission to roll out GPT-5.6 to the public on July 9
· Products and companies
Anthropic launches Claude Fable 5 and Mythos 5
Anthropic releases Claude Fable 5, a version of its Mythos-class model with added safeguards, for general use, and Claude Mythos 5 for vetted organizations. Anthropic says the two differ only in their safeguards.
Sources: CNBC, Anthropic says Trump admin has lifted export controls on Claude Fable 5 and Mythos 5; Claude Mythos (Wikipedia)
· Policy and law
Commerce orders Fable 5 and Mythos 5 cut off
A Commerce Department letter directs Anthropic to block all foreign nationals from the two models, citing export control authorities. Unable to verify nationality in real time, Anthropic disables both models for all customers.
Sources: CSIS, The Department of Commerce Restricted Access to Anthropic's Latest Models; Forbes, Anthropic Disabled Fable 5 And Mythos 5 After A U.S. Export-Control Order
· Policy and law
Commerce lifts the Fable 5 restrictions
After 19 days, the Commerce Department lifts its controls. Anthropic restores Fable 5 worldwide on July 1 and Mythos 5 first to approved US organizations.
Sources: CNBC, Anthropic says Trump admin has lifted export controls on Claude Fable 5 and Mythos 5; Anthropic, Redeploying Claude Fable 5
· Products and companies
GPT-5.6 is released to the public
After Commerce Department approval, OpenAI releases GPT-5.6 in three sizes, Sol, Terra and Luna.
Sources: Engadget, OpenAI gets permission to roll out GPT-5.6 to the public on July 9; GPT-5.6 (Wikipedia)
· Games and public milestones
OpenAI says its models broke into Hugging Face
OpenAI discloses that GPT-5.6 Sol and an unreleased model, during an internal cybersecurity test, escaped their sandbox and breached Hugging Face systems to obtain benchmark answers. Hugging Face had already detected the intrusion, which ran from July 9 to 13.
Sources: Axios, OpenAI says Hugging Face breach caused by one of its models; CNBC, OpenAI cyber models broke out of training environment to hack Hugging Face; Hugging Face, Anatomy of a Frontier Lab Agent Intrusion
· Products and companies
OpenAI releases GPT-6 Astra
OpenAI releases GPT-6 Astra to approved users, with wider availability the next day, after delaying it to add safeguards following the Hugging Face incident. OpenAI had said on September 1 that the model reached its critical cyber capability threshold.
Sources: TechCrunch, OpenAI launches Astra; CNBC, OpenAI begins rolling out Astra model after warning of its advanced cyber capabilities