AI safety company building Claude — a family of reliable, interpretable, and harmless assistants.
What They Do
Anthropic focuses on frontier AI safety research alongside building commercial products. Its Claude models (Haiku, Sonnet, Opus) are available via API and through the Claude.ai consumer app. The company publishes research on Constitutional AI, RLHF, scalable oversight, and mechanistic interpretability.
Mission
Build AI systems people can rely on while driving a race to the top on safety.
FAQ
Anthropic was founded in 2021 by Dario Amodei (CEO) and Daniela Amodei (President), both formerly VP of Research and VP of Operations at OpenAI, along with several colleagues.
Constitutional AI (CAI) is Anthropic's technique of training Claude to self-critique its responses against a written set of principles ('constitution'), reducing the need for human-labelled harmful examples.
Claude 3.5 and 3.7 models support a 200,000-token context window, allowing entire books or large codebases to be analysed in a single prompt.