Loading
Poseidon Research is an independent AI safety lab in New York City. It works on deception and hidden information in frontier AI, and its work spans control, mechanistic interpretability, information theory and AI security. It builds evaluations and open-source tools, and its team has helped maintain TransformerLens, a widely used interpretability framework. It is a registered 501(c)(3) nonprofit, so it sells no products. I could not verify its funding, revenue or headcount.