Transluce is an independent nonprofit AI research lab that builds open, scalable technology for understanding AI systems and steering them in the public interest.
Transluce is an independent nonprofit AI research lab that builds open, scalable technology for understanding AI systems and steering them in the public interest.
People
Updated 05/18/26Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- $11,000,000
- Funding Raised to Date
- -
Org Details
Updated 05/18/26Transluce is an independent nonprofit AI research lab headquartered in San Francisco, California. The organization was co-founded on October 23, 2024 by Jacob Steinhardt, an assistant professor at UC Berkeley's Department of Statistics, and Sarah Schwettmann, a research scientist at MIT CSAIL. The name "Transluce" means to shine light through something to reveal its structure, reflecting the lab's core commitment to transparency. The lab's mission is to build open, scalable technology for understanding AI systems and steering them in the public interest. Transluce believes that as AI systems grow more capable, human oversight must scale alongside them—something that requires automated tools rather than purely manual inspection. Its initial technical releases included automated feature descriptions for neuron activation patterns, an observability interface for interrogating AI system internals, and a behavior elicitation agent capable of searching for specific behaviors in frontier models such as Llama-405B and GPT-4o. Transluce is structured as a 501(c)(3) nonprofit (legal name: Clarity AI Research Inc.) and follows an open-core business model: the core oversight stack is public and open-source, while hosted services and advanced features generate earned revenue. As of 2025, earned revenue from private companies and governments represents approximately 20% of Transluce's funding, with the remainder coming from philanthropic donors including Schmidt Sciences and Halcyon Futures. The organization established the AI Evaluator Forum, a collaborative body that brings together leading researchers to set shared standards for independent AI evaluators, and released AEF-1, a standard for ensuring minimum levels of access, transparency, and independence for third-party evaluations. In 2025 the lab published an end-of-year fundraiser seeking $11 million to expand its research, scale compute infrastructure, conduct public safety audits of openly available models such as DeepSeek and Llama, and support governance and public accountability work. The team has grown to approximately 20 core members, including members of technical staff, research fellows, a senior governance fellow, and operations staff. Notable advisors include Yoshua Bengio (Université de Montréal), Percy Liang (Stanford), and Jacob Andreas (MIT). The board consists of Steinhardt, Alex Allain, and Mike McCormick.
Theory of Change
Updated 05/18/26Transluce believes that scalable democratic oversight of AI requires automated tools that can match the pace and complexity of modern AI development. Their causal chain is: (1) build open-source AI-driven tools that can automatically analyze and explain the internals and behaviors of large AI models; (2) put these tools in the hands of independent evaluators, governments, and civil society so that safety assessments are no longer controlled solely by commercial labs; (3) establish shared industry standards—through bodies like the AI Evaluator Forum—that normalize independent auditing; (4) use public audits and transparency to create accountability pressure that nudges AI developers toward safer deployment practices. By operating as a nonprofit that openly publishes its methods, Transluce aims to become a trusted, independent reference point that can credibly identify risks such as deception, hallucination, and misuse before they cause harm.
Grants Received– no grants recorded
Updated 05/18/26Projects
Updated 05/18/26
A 2025 fundraising campaign to support Transluce in scaling democratic oversight of AI by building automated oversight tools and deploying them to evaluators, companies, governments, and civil society.
Discussion
Key risk: Their impact is bottlenecked by access and adoption: if labs withhold robust evaluator access or their automated interpretability/evaluation methods fail to generalize to deceptive frontier systems, the ambitious, compute-intensive scale-up could have low counterfactual value and risk capability spillovers.
Case for funding: Transluce, led by Jacob Steinhardt and Sarah Schwettmann, is building open, automated oversight tools and evaluator standards (AEF-1) that enable independent audits of frontier models at scale, a high-leverage way to shift industry norms and empower governments and civil society to impose accountability on AI deployment.