Organisation · nonprofit · US

Transluce

A nonprofit building open tools to understand and oversee AI systems, including automated "investigator agents" and interpretability methods.

Reference profile

Transluce is a San Francisco AI-research nonprofit founded in 2024 by Jacob Steinhardt, who took leave from a Berkeley professorship to start it, and Sarah Schwettmann, formerly of MIT. It builds open tools for understanding and overseeing AI systems, describing its aim as a public "tech stack" for scalable oversight. Its early work centred on "investigator agents" — automated systems that probe a target model to surface specific behaviours such as harmful responses or hallucinations — and on interpretability methods for examining model internals. As a nonprofit releasing its work openly, it positions itself as an independent counterpart to the safety teams inside the frontier labs.

Category
Safety & alignment research
Founded
2024
HQ
San Francisco, US