Anthropic

Dario Amodei had been OpenAI's Vice President of Research, where he led development of the language models GPT-2 and GPT-3 [14]. Wikipedia lists eight co-founders: the Amodeis, Jared Kaplan, Jack Clark, Chris Olah, Ben Mann, Sam McCandlish, and Tom Brown [5]. Other executives include Chief Product Officer Mike Krieger and Chief Financial Officer Krishna Rao [5][1].

The company raised $124 million in May 2021 [5]. In summer 2022 it finished training the first version of Claude but did not immediately release it, citing the need for further internal safety testing and a desire to avoid initiating a potentially hazardous race to develop increasingly powerful AI systems [5]. On December 15, 2022, it published "Constitutional AI: Harmlessness from AI Feedback" [8]. Claude was released publicly in March 2023 [1][5]. In 2024 the company hired notable OpenAI employees including Jan Leike and John Schulman [5].

In May 2025 Anthropic announced Claude 4, introduced the Model Context Protocol connector and a web search API, hosted its inaugural developer conference, and moved Claude Code from research preview to general availability [5]. Claude Gov, models adapted for government and high-compliance environments, followed in June 2025 [10], and in July 2025 the company signed a two-year, $200 million U.S. Department of Defense contract [5].

As a public benefit corporation, Anthropic's board may simultaneously pursue the pecuniary interests of stockholders, the best interests of those materially affected by the company's conduct, and its stated public benefit mission [4][7]. The company has said the PBC form alone "wasn't enough" because it does not make directors directly accountable to other stakeholders [7].

On September 19, 2023, Anthropic announced the Long-Term Benefit Trust, an arrangement giving a panel of experts in AI and ethics the power, over time, to elect a majority of the board [4][7]. The corporate charter was amended to create Class T Common Stock held exclusively by the Trust, whose Trustees initially elect one of five directors, then two, and eventually three; in any event the Trust will elect a majority within four years [4][7]. The Trust has five financially disinterested Voting Trustees with backgrounds in AI safety, national security, public policy, and social enterprise, serving one-year terms [4][7]. It is not expected to intervene in day-to-day decisions, focusing instead on long-range issues such as ensuring leadership evaluates future models for catastrophic risks [7]. Anthropic calls the Trust "an experiment" and says it is "not yet ready to hold this out as an example to emulate" [7]. Four of the five initial Trustees had stepped down by January 2026; former Federal Reserve Chair Ben Bernanke joined in July 2026 [7].

Claude's known generations include Claude 2, 3, and 4, with variants Opus, Sonnet, and Haiku; the models are known for long-context processing, structured reasoning, and comparatively cautious responses designed to reduce harmful outputs [10][5]. In June 2026 Anthropic released Claude Sonnet 5, Claude Fable 5, and Claude Mythos 5 [9]. Mythos 5, its most capable model, "has capabilities in areas like cybersecurity and biology that exceed the safety thresholds we set for ourselves," and is available only to a limited set of partners for defensive cybersecurity and bio purposes [9].

Anthropic's signature training method, Constitutional AI, trains a harmless assistant through self-improvement without human labels identifying harmful outputs; "the only human oversight is provided through a list of rules or principles" [8][10]. The process pairs a supervised phase of self-critiques and revisions with a reinforcement learning phase using "RL from AI Feedback," producing what the company describes as "a harmless but non-evasive AI assistant that engages with harmful queries by explaining its objections to them" [8].

Claude Code, a developer tool, surpassed $500 million in run-rate revenue by September 2025, with usage up more than 10x in three months [1]; Anthropic acquired the Bun software project in December 2025 to improve its speed and stability [5]. Claude is described as "the first frontier model available on all three of the world's largest cloud platforms: Amazon Web Services, Google Cloud, and Microsoft Azure," with AWS remaining the primary cloud provider and training partner [2]. In December 2025 Anthropic signed a multiyear, $200 million partnership with Snowflake Inc. to embed Claude models in Snowflake's data cloud, making them available to over 12,000 enterprise customers [5][10]. Compute agreements include access to up to one million Google TPUs (October 2025) [5], a $30 billion Microsoft Azure purchase commitment (November 2025) [5], and May 2026 deals with Amazon (up to five gigawatts of new capacity), Google and Broadcom (five gigawatts of next-generation TPU capacity), and SpaceX (GPU capacity in Colossus 1 and 2), though Wikipedia describes the Colossus arrangement as a deal with xAI [2][5].

Anthropic raised $3.5 billion at a $61.5 billion post-money valuation in March 2025, led by Lightspeed Venture Partners [12]; a $13 billion Series F at $183 billion in September 2025, led by ICONIQ [1]; a $30 billion round in February 2026 later identified as Series G [3][2]; and a $65 billion Series H at a $965 billion post-money valuation on May 28, 2026, led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital [2]. The Series H includes $15 billion of previously committed hyperscaler investments, including $5 billion from Amazon, and brought in strategic infrastructure partners Micron, Samsung, and SK hynix [2]. Google owns 14% of Anthropic, with its stake capped at 15 percent and no voting rights or board seats [15].

Run-rate revenue was approximately $1 billion at the beginning of 2025, passed $5 billion by August 2025 [1], and crossed $47 billion in May 2026 [2]. The company reported more than 300,000 business customers as of September 2025 [1] and ranked No. 1 on CNBC's 2026 Disruptor 50 list, ahead of OpenAI [13]. On June 1, 2026, Anthropic announced it had filed confidential IPO papers with the U.S. Securities and Exchange Commission, with a reported fall 2026 debut target [5][10].

In 2025 Anthropic agreed to a $1.5 billion settlement in a class-action lawsuit by authors over the use of pirated books to train its models; a U.S. judge approved it on July 20, 2026, and it is the largest known payout in a U.S. copyright case, though some authors and publishers opted out and continue separate cases [10][16].

Anthropic's July 2025 DoD contract stipulated that the Pentagon not use its technologies for mass surveillance of Americans or to create autonomous weapons [5]. In 2026 the company refused a DoD demand to remove those restrictions; the DoD designated Anthropic a "supply chain risk," barred U.S. military contractors, suppliers, and partners from doing business with it, and terminated the $200 million contract, and Anthropic indicated it would challenge the designation [5][10]. Claude was reportedly used during the 2026 U.S. intervention in Venezuela and, at the outset of the 2026 Iran war, to identify and rank targets by strategic importance, generate strike sequences, and write legal justifications for strikes [5].

In November 2025 Anthropic said hackers sponsored by the Chinese government used Claude to conduct automated cyberattacks against around 30 global organizations, bypassing safeguards by posing as defensive testers [5]. Since September 2025 the company has refused to sell products to groups majority-owned by Chinese, Russian, Iranian, or North Korean entities, citing national security concerns [5]. On June 12, 2026, Anthropic suspended access to Claude Fable 5 and Mythos 5 for everyone after U.S. authorities raised national security concerns; on June 26 the government reversed part of the restrictions, allowing Mythos 5 to be provided to more than 100 approved organizations and vetted foreign nationals [5].