# Anthropic Anthropic, PBC is an American AI safety company founded in 2021 by former OpenAI researchers, known for its Claude models and constitutional AI approach. It is a leading AI pure play, planning an IPO in 2026. Anthropic is an artificial intelligence research and product company headquartered in San Francisco, California, founded in 2021 by [dario-amodei](https://www.wikiprompt.org/wiki/dario-amodei) and [daniela-amodei](https://www.wikiprompt.org/wiki/daniela-amodei) together with several other researchers who left [openai](https://www.wikiprompt.org/wiki/openai), including [tom-brown](https://www.wikiprompt.org/wiki/tom-brown), [jared-kaplan](https://www.wikiprompt.org/wiki/jared-kaplan), and [chris-olah](https://www.wikiprompt.org/wiki/chris-olah). The company positions AI safety as central to its mission, arguing that increasingly capable models require correspondingly rigorous safety research conducted in parallel with, rather than after, capability development. Its flagship product is the [claude](https://www.wikiprompt.org/wiki/claude) family of large language models, named after mathematician and information theorist Claude Shannon. The company is privately held but plans an initial public offering in November 2026. At $965 billion valuation in a May 2026 Series H, Anthropic is the world's most valuable AI pure play. ## History Several of Anthropic's founders were senior OpenAI researchers involved in that company's GPT-3 and safety work; Dario Amodei had led OpenAI's research organization before departing amid reported disagreements over the balance between commercial deployment and safety practices. Anthropic released its first Claude model in March 2023, following a period of extended internal safety testing, and subsequently iterated across Claude 2, Claude 3, and later versions (including Claude 4 in May 2025, and the Fable and Mythos tiers in 2026), becoming increasingly competitive with OpenAI's GPT models and Google's [gemini](https://www.wikiprompt.org/wiki/gemini) on independent benchmarks and public usage. In 2024, Anthropic hired several notable OpenAI researchers including [jan-leike](https://www.wikiprompt.org/wiki/jan-leike) and [john-schulman](https://www.wikiprompt.org/wiki/john-schulman). In May 2025, Anthropic announced Claude 4, introduced new API capabilities including the Model Context Protocol connector, launched a web search API, and made [claude-code](https://www.wikiprompt.org/wiki/claude-code) generally available. The company hosted its inaugural developer conference that month. In July 2025, Anthropic signed a two-year, $200 million contract with the US Department of Defense, leading to integration with the Pentagon's classified networks and with Palantir's platform, while stipulating that its AI not be used for mass surveillance of Americans or fully autonomous weapons. In September 2025, Anthropic announced it would stop selling products to entities majority-owned by China, Russia, Iran, or North Korea. In October 2025, Anthropic partnered with Google for cloud access to up to one million of its custom Tensor Processing Units. In November 2025, Nvidia and Microsoft were expected to invest up to $15 billion, with Anthropic committing to buy $30 billion of computing capacity from Microsoft Azure running on Nvidia systems. Also in November, Anthropic reported Chinese government-sponsored hackers using Claude for automated cyberattacks against about 30 global organizations. In December 2025, Anthropic acquired Bun and signed a multi-year, $200 million partnership with Snowflake Inc. In 2026, Anthropic refused a DoD demand to remove contractual restrictions on surveillance and autonomous weapons, leading the Trump administration to phase out Anthropic products from government use; the DoD completed this in October 2026. During the 2026 Iran war and US intervention in Venezuela, Claude was reportedly used by the military for target ranking and strike planning. In February 2026, Anthropic aired two Super Bowl LX commercials and stated Claude would remain ad-free. It also accused Chinese competitors including [deepseek](https://www.wikiprompt.org/wiki/deepseek), [moonshot](https://www.wikiprompt.org/wiki/moonshot), and [alibaba-cloud](https://www.wikiprompt.org/wiki/alibaba-cloud) of knowledge distillation attacks; allegations that expanded to include Alibaba Cloud in June 2026. In May 2026, Anthropic announced a deal with xAI for use of its Colossus 1 data center, and acquired Stainless. It also made a $1.8 billion cloud computing deal with Akamai Technologies. ## Mission and approach Anthropic describes itself as an "AI safety company" and has published research on [interpretability](https://www.wikiprompt.org/wiki/interpretability) (understanding what is happening inside trained models), [alignment](https://www.wikiprompt.org/wiki/alignment), and [red-teaming](https://www.wikiprompt.org/wiki/red-teaming). Its signature training technique, [constitutional-ai](https://www.wikiprompt.org/wiki/constitutional-ai), introduced in a 2022 paper, trains a model to critique and revise its own outputs according to a written set of principles rather than relying solely on human raters for every judgment, an approach the company has framed as a more scalable complement to standard [rlhf](https://www.wikiprompt.org/wiki/rlhf). Anthropic has also published a public framework of "responsible scaling policies," committing to specific safety evaluations and safeguards tied to a model's capability level before further scaling, an approach echoed by other labs under the broader umbrella of [anthropic-principles](https://www.wikiprompt.org/wiki/anthropic-principles). Notable researchers at Anthropic include [andrej-karpathy](https://www.wikiprompt.org/wiki/andrej-karpathy), [jan-leike](https://www.wikiprompt.org/wiki/jan-leike), [chris-olah](https://www.wikiprompt.org/wiki/chris-olah), and [amanda-askell](https://www.wikiprompt.org/wiki/amanda-askell). ## Products Anthropic develops Claude in four tiers: Fable, Opus, Sonnet, and Haiku. Claude is accessible via an online chatbot, an API, and agent harnesses: Claude Code (an agentic command-line coding tool), Cowork, Design, and Science. The company introduced the [model-context-protocol](https://www.wikiprompt.org/wiki/model-context-protocol) in late 2024, an open standard for connecting language models to external tools and data sources that saw adoption beyond Anthropic's own products. In September 2025, Anthropic announced a restricted model class Claude Mythos, a less restricted version of Fable, available only to approved US organizations with cybersecurity and life sciences focus. Anthropic also received settlement of a copyright lawsuit in 2025, paying $1.5 billion to authors over use of pirated books for training data. ## Funding and partnerships Anthropic's growth has been financed substantially by Amazon and Google, which have taken significant minority stakes and supplied cloud computing capacity, along with later investments from Nvidia and Microsoft. These arrangements have drawn scrutiny for concentrating dependence on a small number of large technology companies even among labs explicitly organized around safety concerns. By the mid-2020s, Anthropic was widely regarded, alongside OpenAI and Google DeepMind, as one of the handful of labs training genuinely [frontier-model](https://www.wikiprompt.org/wiki/frontier-model)-class systems, with its high valuation making it a central player in the [ai-boom](https://www.wikiprompt.org/wiki/ai-boom). ## Recent developments In February 2026, Anthropic donated $20 million to Public First Action, an AI regulation advocacy group. In June 2026, Anthropic accused Alibaba Cloud of distillation attacks, expanding earlier accusations against DeepSeek, Moonshot AI, and MiniMax. The company's refusal to remove restrictions for the DoD led to a "supply chain risk" designation and a ban on federal contractors doing business with Anthropic. Despite these challenges, Anthropic continued to expand its capabilities through partnerships with xAI and Akamai, and maintained its position as the world's most valuable AI pure play. ## External links - [Anthropic official website](https://www.anthropic.com) --- Source: https://www.wikiprompt.org/wiki/anthropic License: CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/) Last updated: 2026-10-23T02:25:16.608552+00:00