Science of Trustworthy AI

AI Interpretability RFP

Funds mechanistic interpretability research on detecting and steering deceptive behaviors in artificial intelligence systems.

⚠ This may reflect a past cycle — verify the current call on the funder's site.

The 2026 AI Interpretability RFP, issued by Schmidt Sciences under its Science of Trustworthy AI program family, funded research on detecting and correcting deceptive behaviors in AI systems. Schmidt Sciences defined deceptive behaviors to include factually incorrect statements, misleading confidence claims, fabrications, selective omission, evasiveness, and false claims regarding a model's self-knowledge. Three research directions were in scope: Detection — tools to identify cases where a model's output contradicts its internal representations; Steering — targeted interventions using interpretability insights to improve model truthfulness; and Applications — translating detection and steering methods into practical human-AI collaboration and multi-agent systems.

Awards ranged from $300,000 to $1,000,000 inclusive of overhead, with indirect costs capped at 10% of the total award. Project duration was 1 to 3 years, with applicants proposing their own timeline within that range. Eligibility was global and open to individual researchers, research teams, universities, national laboratories, institutes, and nonprofits. For-profit entities and organizations conducting lobbying or political activities were excluded. The application window ran from March 16, 2026 through May 26, 2026 at 11:59 PM Anywhere on Earth, with informational webinars held on April 2 and April 28, 2026. Decision notifications were expected in Summer 2026.

Applications were submitted exclusively through the SurveyMonkey Apply portal at schmidtsciences.smapply.io. Competitive proposals require grounding in mechanistic interpretability methods and must connect directly to one or more of the three defined research directions; proposals that are only tangentially related to deception detection or steering are unlikely to be funded. Researchers with questions may contact interpretability@schmidtsciences.org. This RFP is a sibling to the broader Trustworthy AI RFP (deadline May 17, 2026) and forms part of Schmidt Sciences' sustained AI safety research program.