What Is Anthropic? Inside the AI Safety Company Behind Claude

A Company Born From a Disagreement

Anthropic PBC is an American artificial intelligence company headquartered in San Francisco, California, best known for developing the Claude family of large language models. Unlike most technology companies, Anthropic's origin story is not built around a market opportunity or a clever piece of code. It is built around a disagreement about risk.

The company was founded in January 2021 by seven former employees of OpenAI, including the siblings Daniela Amodei and Dario Amodei. Dario Amodei had served as OpenAI's Vice President of Research and was deeply involved in the development of GPT-2 and GPT-3, the systems that demonstrated just how quickly language models could scale in capability. According to multiple accounts of the split, Amodei and his colleagues came to believe that the pace of capability scaling at OpenAI was outrunning the corresponding investment in safety research. Rather than continuing to push from the inside, fourteen to fifteen colleagues left over the course of a few weeks in late 2020 and early 2021 to start something new.

Choosing a Name

The team considered a number of names before settling on Anthropic, a word chosen partly because it evoked being human-centered and partly because the domain name happened to be available. The company raised $124 million in its initial funding round in May 2021, a relatively modest sum by the standards of what would later become one of the most heavily capitalized companies in technology history.

A Public Benefit Corporation

Anthropic was deliberately structured as a public benefit corporation, a legal form that requires a company to balance shareholder returns against a stated public mission. In Anthropic's case, that mission centers on ensuring that increasingly powerful AI systems remain safe, interpretable, and steerable as they scale. This structure is meant to give the company's board legal cover to make decisions that might not maximize short-term profit if those decisions serve the company's safety mission — a governance choice that has shaped many of Anthropic's public stances, including disputes over how its models may be used.

From Research Lab to Major Commercial Force

In the summer of 2022, Anthropic finished training the first version of its Claude model but chose not to release it immediately, citing further internal safety testing. Claude eventually launched the following year, and the company's commercial trajectory since then has been extraordinary. Anthropic's annualized revenue run rate grew from roughly $1 billion in early 2025 to tens of billions within about a year, driven heavily by enterprise and developer adoption of its API and coding-focused tools. By 2026, independent valuations placed the company among the most valuable privately held technology companies in the world.

Research Contributions Beyond Claude

Anthropic has also become known for interpretability research — the attempt to understand what is actually happening inside a large language model rather than treating it as a black box. Using a technique called dictionary learning, Anthropic researchers identified millions of internal patterns inside Claude models, including features tied to specific, identifiable concepts. This research is part of a broader effort the company calls mechanistic interpretability, which aims to make advanced AI systems less opaque to the humans building and overseeing them.

Why It Matters

Anthropic's significance lies less in any single product and more in the position it occupies in the AI industry: a frontier AI lab built from the start around the premise that capability and safety research need to advance together, not sequentially. Whether that bet pays off in the way its founders hoped remains an open question that will likely take years to resolve, but it has undeniably shaped how the broader AI industry talks about safety, alignment, and responsible scaling.