Anthropic is founded

Anthropic is founded

“Anthropic, maker of Claude, was founded in 2021 by Dario and Daniela Amodei and other ex-OpenAI staff as a public-benefit company that put AI safety first; Amazon and Google money made it one of the biggest AI labs.”

Anthropic is an American artificial-intelligence company, best known for the Claude models. It began when a group of researchers left OpenAI at the end of 2020; OpenAI announced on 29 December 2020 that its vice-president of research, Dario Amodei, was leaving with 'a handful' of colleagues to start a new project. Anthropic was formed in Delaware on 26 January 2021 as a public-benefit corporation: a company that makes profits for its shareholders but whose directors must also weigh a stated public purpose, here 'the responsible development and maintenance of advanced AI for the long-term benefit of humanity'. Dario Amodei became chief executive and his sister Daniela Amodei president. The company names five more co-founders: Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Chris Olah. Several had worked on OpenAI's GPT-2 and GPT-3 models and on 'scaling laws', the finding that language models improve steadily as they get more data and computing power.

The founders said they would study how to make large AI systems safer and easier to understand, but that this required building such systems themselves, at great cost. The money came fast. A first round of $124 million (May 2021) was followed by $580 million in April 2022, most of it from Sam Bankman-Fried's FTX group; after FTX collapsed, its bankruptcy estate sold most of that stake in 2024 for $884 million. Claude was released on 14 March 2023, the same day as OpenAI's GPT-4. From 2023 the two largest cloud companies became both investors and suppliers: Amazon put $8 billion into Anthropic in convertible notes between 2023 and 2025 and became its main cloud and chip provider, and Google invested and rented it its TPU chips. Regulators in the United States and Britain examined these ties; Britain's competition authority decided in 2024 that they were not mergers. By May 2026 Anthropic said investors valued it at $965 billion, and in June 2026 it filed confidentially for a stock-market listing.

Anthropic presented safety as its reason to exist. In September 2023 it published a Responsible Scaling Policy, which ties stricter safeguards to more dangerous capabilities, and set up a Long-Term Benefit Trust that elects part of its board. Supporters saw these as models for the industry; critics answered that the rules were voluntary, written and judged by the company itself, and that a firm warning about the risks of AI while racing to build stronger systems feeds the race it warns about. In February 2026 the company rewrote the policy, separating what it would do alone from what it recommended for the whole industry and adding public but non-binding goals; it said the strictest safeguards might be impossible for one company to meet alone. Its own reports also showed the risks: in November 2025 it described a cyber-espionage campaign, which it attributed to a Chinese state-sponsored group, that had used its coding tool against about thirty targets.

Courts became a large part of the story. Authors and music publishers sued Anthropic for training on their work without permission. In June 2025 Judge William Alsup ruled that training on books was a fair use, but that downloading more than seven million pirated books to build a library was not protected; Anthropic then agreed to pay $1.5 billion, about $3,000 for each of some 482,000 books, which the court called 'the largest copyright class action settlement in history'; it was approved in July 2026. In early 2026 Anthropic refused the Pentagon's demand to allow 'any lawful use' of Claude, keeping bans on mass domestic surveillance and fully autonomous weapons. The government labelled it a supply-chain risk and ordered federal agencies to stop using its products. A federal judge in San Francisco found this to be unlawful retaliation, while an appeals court in Washington upheld the Defense Department's narrower exclusion of Claude from its own systems.

Anthropic is one of several laboratories, alongside OpenAI, Google DeepMind, Meta, xAI and Chinese groups such as DeepSeek, that drove the AI boom of the 2020s. Its history shows the economics of that boom: building frontier models needed so much computing power that even a company founded to be cautious became dependent on Amazon, Google and, later, Microsoft and Nvidia. It also shows the open questions of the period: whether private companies can govern a powerful technology through voluntary rules, how copyright applies to machine learning, and how far a government can direct the companies that build AI. As of October 2026 Anthropic is still private and these disputes are not settled.

What happened

  1. 2020

  2. OpenAI announces that Dario Amodei, its vice-president of research, is leaving after nearly five years; with 'a handful' of colleagues he plans a new project that will focus 'less on product development and more on research'.

  3. 2021

  4. Anthropic, PBC is formed in Delaware as a public-benefit corporation; a federal judge later describes it as founded by former OpenAI employees in January 2021.

  5. Anthropic announces its first funding, $124 million led by Jaan Tallinn, a co-founder of Skype; Dario Amodei is chief executive and Daniela Amodei president.

  6. Co-founder Ben Mann downloads at least five million books from Library Genesis, which, in the court's later summary of the record, he knew to be pirated; they go into the company's book library.

  7. Alameda Research Ventures, part of Sam Bankman-Fried's FTX group, pays $500 million for a right to future Anthropic shares.

  8. 2022

  9. Anthropic announces a $580 million Series B led by Bankman-Fried; when it closes the next day, the FTX group owns about 13.6 percent of the company.

  10. Anthropic researchers publish 'Constitutional AI: Harmlessness from AI Feedback', a way to train a model against a written list of principles with AI feedback instead of human ratings.

  11. 2023

  12. Anthropic names Google Cloud its cloud provider for training and running its models.

  13. Claude and the faster Claude Instant are released through a chat interface and an API, after a closed test with Notion, Quora and DuckDuckGo; OpenAI releases GPT-4 the same day.

  14. Vice-President Harris meets the chief executives of Alphabet, Anthropic, Microsoft and OpenAI at the White House; President Biden drops by to say the companies must make their products safe before release.

    The White House, Washington: the meeting with Vice-President Harris (4 May 2023) and the voluntary commitments (21 July 2023)

  15. A $450 million Series C led by Spark Capital includes Google among the investors.

  16. Prime Minister Rishi Sunak receives Dario Amodei, Sam Altman of OpenAI and Demis Hassabis of Google DeepMind at Downing Street.

    10 Downing Street, London: Rishi Sunak meets the heads of Anthropic, OpenAI and Google DeepMind (24 May 2023)

  17. Claude 2 is released, with claude.ai, a public website for chatting with it.

  18. Anthropic is one of seven companies whose voluntary safety commitments President Biden announces at the White House.

    The White House, Washington: the meeting with Vice-President Harris (4 May 2023) and the voluntary commitments (21 July 2023)
  19. 3:00 p.m.

    Dario Amodei testifies with Yoshua Bengio and Stuart Russell before a Senate Judiciary subcommittee on principles for regulating AI.

  20. Anthropic publishes its Responsible Scaling Policy, with 'AI Safety Levels' that require stronger safeguards as models grow more capable, and describes its Long-Term Benefit Trust.

  21. Amazon agrees to invest up to $4 billion, starting with a $1.25 billion convertible note; Amazon Web Services becomes Anthropic's main cloud provider for critical work, with Amazon's Trainium chips.

  22. Music publishers led by Concord, Universal and ABKCO sue Anthropic in Nashville, saying Claude was trained on and reproduces song lyrics.

  23. At the AI Safety Summit at Bletchley Park, Dario Amodei presents the Responsible Scaling Policy to governments.

    Bletchley Park: the AI Safety Summit, where Anthropic presented its Responsible Scaling Policy (1 November 2023)
  24. 2024

  25. The US Federal Trade Commission orders Alphabet, Amazon, Anthropic, Microsoft and OpenAI to report on the big companies' investments in AI start-ups.

  26. The Claude 3 models, Haiku, Sonnet and Opus, are released.

  27. The Delaware bankruptcy court approves the sale of most of the FTX estate's Anthropic shares, 29.5 million of them, for $884 million; an Abu Dhabi fund, ATIC, takes the largest part.

  28. The authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sue Anthropic in San Francisco for training Claude on pirated copies of their books.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  29. Britain's Competition and Markets Authority decides that Amazon's partnership with Anthropic does not qualify for a merger investigation; it reaches the same view on Google's on 19 November.

  30. Britain's Science Secretary Peter Kyle visits Anthropic's San Francisco offices.

  31. Amazon invests another $4 billion, bringing its total to $8 billion, and AWS becomes Anthropic's primary partner for training as well as cloud.

  32. 2025

  33. Anthropic releases Claude 3.7 Sonnet and a preview of Claude Code, a tool that writes and runs software on a programmer's computer.

  34. With Claude Opus 4 Anthropic switches on its stricter 'ASL-3' safeguards for the first time, as a precaution against help with biological and chemical weapons; Claude Code becomes generally available.

  35. Judge William Alsup rules that training Claude on books was a fair use, but that keeping more than seven million pirated books in a central library was not, and orders a trial on the piracy.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  36. The Defense Department's AI office awards Anthropic a prototype agreement worth up to $200 million.

  37. Anthropic changes its consumer terms: users of the free and paid Claude apps must choose whether their chats may be used to train future models.

  38. The authors ask the court to approve a $1.5 billion settlement, about $3,000 per book, under which Anthropic will also destroy its pirated files.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  39. Anthropic agrees to use up to one million Google TPU chips, a deal it values at tens of billions of dollars.

  40. Dario Amodei meets Prime Minister Sanae Takaichi in Tokyo as Anthropic opens its first Asian office.

    The Prime Minister's Office, Tokyo: Dario Amodei meets Prime Minister Takaichi as Anthropic opens its Tokyo office (29 October 2025)

  41. Anthropic reports that a group it attributes to the Chinese state used Claude Code in an espionage campaign against about thirty organisations, succeeding in a few cases.

  42. Microsoft and Nvidia agree to invest up to $5 billion and $10 billion; Anthropic commits to buy $30 billion of computing power on Microsoft's Azure.

  43. 2026

  44. A $30 billion Series G values Anthropic at $380 billion.

  45. Dario Amodei is among the technology chiefs with Prime Minister Modi at the India AI Impact Summit in New Delhi.

    Bharat Mandapam, New Delhi: the India AI Impact Summit (19 February 2026)

  46. Version 3 of the Responsible Scaling Policy separates the company's own commitments from its recommendations for the industry and adds public but non-binding goals; Anthropic writes that parts of its original plan 'have not panned out'.

  47. Dario Amodei refuses the Pentagon's demand that Claude be available for 'any lawful use', keeping two exceptions: mass domestic surveillance of Americans and fully autonomous weapons.

    The Pentagon: the Department of War, which designated Anthropic a supply-chain risk in 2026

  48. President Trump orders federal agencies to stop using Anthropic's products and Secretary of War Pete Hegseth directs that the company be designated a supply-chain risk, a label that Anthropic says had never before been applied in public to an American company.

    The Pentagon: the Department of War, which designated Anthropic a supply-chain risk in 2026

  49. Anthropic sues the government in San Francisco and petitions the Court of Appeals in Washington to overturn the designation.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  50. Judge Rita F. Lin grants Anthropic a preliminary injunction against the government's measures; the government appeals.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  51. Anthropic commits more than $100 billion over ten years to Amazon's cloud for up to five gigawatts of capacity; Amazon adds equity and a facility of up to $20 billion tied to the compute it delivers.

    St. Joseph County, Indiana: one of the Project Rainier data-centre sites Amazon built for Anthropic

  52. A $65 billion Series H values Anthropic at $965 billion; the company puts its annualised revenue above $47 billion.

  53. Anthropic confidentially submits a draft registration statement to the SEC for a possible stock-market listing.

  54. An export-control directive from the US government forces Anthropic to suspend its newest models, Fable 5 and Mythos 5, for all customers; access is restored on 1 July after the controls are lifted.

  55. Judge Araceli Martinez-Olguin gives final approval to the $1.5 billion authors' settlement; objectors appeal in August.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  56. Judge Lin rules that the government's measures were unlawful retaliation for Anthropic's views and denied it due process, and permanently blocks them.

    Phillip Burton Federal Building, San Francisco: the federal court of Bartz v. Anthropic and of Anthropic v. Department of War

  57. The D.C. Circuit, 2 to 1, upholds the Defense Department's exclusion of Claude from its own supply chain under the 2018 supply-chain security law, finding it based on Anthropic's refusal of a contract term, not its views.

    E. Barrett Prettyman Courthouse, Washington: the D.C. Circuit, which upheld the exclusion of Claude (25 September 2026)