2026 Interpretability RFP
Schmidt Sciences funds research to develop and apply interpretability tools that detect and reduce deceptive behaviors in large language models, aiming to make AI systems more truthful and reliable.
Funder: Schmidt Sciences
Due Dates: May 26, 2026 (Full proposal deadline, 11:59pm AoE)
Funding Amounts: $300,000–$1,000,000 per project; project duration 1–3 years
Summary: Funds research to detect and mitigate deceptive behaviors in AI models, aiming for practical, generalizable interpretability tools that enhance AI truthfulness and reliability.
Key Information: Indirect costs capped at 10%; open to global nonprofit research institutions and universities.
Description
This opportunity from Schmidt Sciences supports innovative research to advance interpretability methods for large language models (LLMs), with a focus on detecting and mitigating deceptive behaviors. The program seeks proposals that move beyond academic benchmarks to address real-world risks, developing tools that can identify when AI models provide misleading or harmful advice and interventions that improve model truthfulness. The RFP emphasizes actionable scientific progress, mechanistic insights, and generalizable solutions that enhance the reliability and utility of AI systems. Projects may involve monitoring model reasoning, developing steering methods, and applying these insights to practical AI deployments.
