Anthropic was founded in 2021 by former OpenAI researchers.
Claude
Its main product is the Claude family of language models.
Interpretability
Anthropic invests heavily in mechanistic interpretability — the study of what individual circuits inside a neural network actually compute.