An AI safety company founded in 2021, known for developing the Claude family of large language models with a focus on AI safety and alignment research.
Definition
Anthropic is an AI safety and research company founded in 2021 by former OpenAI researchers, including the siblings Dario and Daniela Amodei. It develops the Claude family of large language models and conducts research into AI safety, alignment, and interpretability. The company operates as a public benefit corporation and has raised substantial venture funding.
Anthropic's generally available frontier model is Claude Opus 4.8, announced in May 2026. During 2026 the company also developed a more capable "Mythos-class" tier oriented towards cybersecurity and scientific research. Claude Mythos Preview, a model with advanced vulnerability-detection capabilities, was distributed through Project Glasswing, a gated programme limited to vetted infrastructure providers, open-source maintainers, and government partners, with use restricted to cybersecurity. Its existence became public in late March 2026 after draft material was inadvertently left in a publicly accessible cache, prompting wide debate about the security implications of highly capable cyber models.
On 9 June 2026 Anthropic launched two further models. Claude Fable 5 is a Mythos-class model intended for general use, in which queries on sensitive domains such as cybersecurity are routed instead to the less capable Claude Opus 4.8. Claude Mythos 5 is the same underlying model with certain safeguards removed, offered to a small group of cyberdefenders through Project Glasswing.
On 12 June 2026, citing national security authorities, the US Department of Commerce issued an export control directive requiring Anthropic to suspend access to Fable 5 and Mythos 5 for any foreign national, whether inside or outside the United States, including the company's own non-citizen employees. Because it could not reliably verify nationality at scale, Anthropic disabled both models for all customers worldwide; access to its other models, including Claude Opus 4.8, was unaffected. Anthropic stated that it understood the directive to follow the discovery of a method to bypass Fable 5's safeguards, described the action as a likely misunderstanding, and said it was working to restore access. From 26 June 2026 the company began restoring Mythos 5 access for US organisations that operate and defend critical infrastructure, while Fable 5 remained unavailable for general use.
Litigation
Anthropic is a defendant in a series of US copyright actions concerning the material used to train Claude, most before the Northern District of California. These include Bartz v. Anthropic PBC (No. 3:24-cv-05417), recorded as settled; Reddit, Inc. v. Anthropic PBC (No. 3:25-cv-05643); Concord Music Group, Inc. v. Anthropic PBC (No. 5:26-cv-00880); BMG Rights Management (US) LLC v. Anthropic PBC (No. 5:26-cv-02334); and Carreyrou v. Anthropic PBC (No. 3:25-cv-10897), each currently active. Anthropic is also among the defendants in Chicken Soup for the Soul, LLC v. Meta Platforms Inc. (No. 5:26-cv-02333). Separately, in Anthropic PBC v. U.S. Department of War (No. 3:26-cv-01996), Anthropic appears as the Plaintiff in a government and regulatory matter.