Anthropic

An American AI safety company founded in 2021 by Dario Amodei, Daniela Amodei, and other former OpenAI researchers, focused on building AI systems that are safe, beneficial, and interpretable. Anthropic develops the Claude family of large language models, designed with an emphasis on helpfulness, harmlessness, and honesty. The company's research contributions include Constitutional AI, a training methodology that uses a written set of principles and a self-critique and revision loop to align model behavior, reducing reliance on costly human preference labeling for every decision. Anthropic has also published influential work in mechanistic interpretability, attempting to understand the internal representations and circuits inside neural networks. It has raised billions in funding from Google, Spark Capital, and others, and is widely regarded as one of the two leading frontier AI labs globally alongside OpenAI, competing closely on model capability benchmarks while emphasizing safety-focused research as a company differentiator. Example: Claude 3 Opus, released in March 2024, became the first publicly available model to outperform GPT-4 on several standard capability benchmarks, establishing Anthropic's position at the frontier of AI development. The Claude 3 family introduced differentiated model tiers (Haiku, Sonnet, Opus) to balance cost and performance, a product structure that has continued through subsequent model generations. Why it matters for AI: Anthropic's influence extends beyond its products into AI governance and safety research. Its Constitutional AI methodology, interpretability research program, and public advocacy for measured AI development have shaped industry practices and regulatory discussions in the US and internationally. It occupies an unusual position as a commercially competitive frontier lab that also publishes substantive work on making AI systems safer and more understandable.

Explore the full Web3 Glossary — 2,000+ expert-curated definitions. Need guidance? Talk to our consultants.