Transluce is an independent, nonprofit research lab that builds open, scalable technology for understanding AI systems and steering them in the public interest. Since launching in October 2024, we've built Docent, our agent evaluation platform used by over 25 organizations including frontier AI labs (Anthropic, DeepMind), third party safety orgs (METR, Redwood, Apollo), government evaluators, and academic labs. Docent was used as part of Claude 4's pre-deployment safety analysis and is integrated with SWE-bench. We build scalable systems for monitoring AI agents, test AI behaviors, interpret AI system inner workings, and study issues like sycophancy, self-harm, and reward hacking. Our open core model keeps our oversight stack public and open source while hosted services generate revenue to subsidize public interest work.