Science of Trustworthy AI↗

Applications: Closed
All 59 grants ↓
Programme of Schmidt Sciences·Founded 2024·New York, NY, United States
“Advancing the Fundamental Science of Trustworthy AI”In its own words: schmidtsciences.org
TechnicalCompute

A Schmidt Sciences grant programme for technical research on understanding, predicting and controlling risks from frontier AI. Its 2026 call offered up to $1M (Tier 1) or $1M-$5M+ (Tier 2) per project, open worldwide.

Show detailsHide details

Schmidt Sciences funds basic technical research on understanding, predicting and controlling risks from frontier AI through open requests for proposals. Staff and external reviewers judge proposals; it does not accept unsolicited proposals.

Typical grant $250k ($250k–$370k)

On applying “We proactively seek out grant opportunities and do not accept unsolicited proposals.” schmidtsciences.org

Itemised here $16M in 59 grants

’24
’25
Led by Stuart Feldman (President)Its grants page ↗Checked 29 Oct 2026

Money flows

Where its money came from and where it went. Pick years and what to show; hover a box or ribbon for details, click one to focus it, click the background to see everything.

How to apply

Applications
Closed
Possibly early 2027 (unconfirmed)
Who can apply
Academics, Organisations
Decision time
About 2–3 months

Open RFP; the 2026 round closed 17 May 2026 with decisions expected in summer 2026.

Programmes

Biggest recipients

All grants

59 grants totalling $16M, newest first. Recipients link to their own pages; people who received grants without a public profile are shown by number.

2025 56 grants · $16M
DateRecipientAmountPurpose
Apr 2025Cybench (Individual #8739 et al.)$50kSafeBench competition first prize ($50,000): cyber-capability benchmark for LM agents. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025AgentDojo (Edoardo Debenedetti et al.)$50kSafeBench competition first prize ($50,000): prompt-injection attacks and defences for LLM agents. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025BackdoorLLM (Yige Li et al.)$50kSafeBench competition first prize ($50,000): backdoor attacks on LLMs. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025CVE-Bench (Yuxuan Zhu et al.)$20kSafeBench competition second prize ($20,000): AI agents exploiting real web vulnerabilities. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025JailBreakV (Weidi Luo et al.)$20kSafeBench competition second prize ($20,000): jailbreak robustness of multimodal LLMs. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025Poser (Individual #2654, Caden Juang, Severin Field)$20kSafeBench competition second prize ($20,000): detecting alignment-faking LLMs. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025Situational Awareness Dataset (Rudolf Laine et al.)$20kSafeBench competition second prize ($20,000): situational awareness in LLMs. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
Apr 2025BioLP-bench (Individual #3015)$20kSafeBench competition second prize ($20,000): biology lab-protocol capability benchmark. Prize pool $250,000 sponsored by Schmidt Sciences, run by the Center for AI Safety; winners page undated, AgentDojo announced its win on 29 Apr 2025. source
11 Feb 2025Zhiqiang Lin$500kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Evaluating and Defending Multimodal Computer-Use Agents under Adversarial Attacks with a Realistic Online Environment. PI identified by Caroline. — Science of Trustworthy AI — AI Safety Science (Feb 2025 co source
11 Feb 2025Sanjeev Arora~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Tests of Compositional Generalization as an 'Upper Bound' on AI Safety Risks. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed (RFP cap ~$500K). — Science of Trustworthy AI source
11 Feb 2025Florian Tramèr~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): ETH Zurich. Project: Formalizing Prompt Injection Defenses. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Zhijing Jin~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Mechanistic Interpretability to Detect Test Set Contamination. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 co source
11 Feb 2025Evan Miyazono and Alexandre Rademaker~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Atlas Computing. Project: Formal Verification of Software. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Ziang Xiao~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Johns Hopkins University. Project: BenchCraft: An AI-powered Toolkit to Create Valid and Efficient Benchmarks with Measurement Theory and Human-AI Collaboration. One of 27 AI Safety Science awards; per-project amoun source
11 Feb 2025Diyi Yang~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Quantifying and Mitigating Privacy Risks in Multi-Agent Systems. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Mykel Kochenderfer~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Stanford University. Project: Adaptive Stress Testing for Automated Unsupervised Large Language Model Testing and Failure Classification. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science source
11 Feb 2025Daniel Kang~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): University of Illinois Urbana-Champaign. Project: Benchmarks for AI Agents and Cybersecurity. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb source
11 Feb 2025Aditi Raghunathan~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Multi-Agent AI Safety via Dynamic Games. PI identified by Caroline. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Yarin Gal~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Tracing and Eliminating Harmful Capabilities Across Model Generations. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Anna Leshinskaya~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Mechanistic Cognitive Alignment for Morally Guided Decision-Making. PI name rendered ambiguously on page; verify before ingest. One of 27 AI Safety Science awards; per-project amount undisclosed. PI identif source
11 Feb 2025Bo Li~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): University of Illinois Urbana-Champaign. Project: Multiagent-Based T&E Environment Construction. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (F source
11 Feb 2025Shlomo Zilberstein~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Unlocking Multi-Agent System Safety with Dynamic Islands of Trust. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 202 source
11 Feb 2025Zico Kolter~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Carnegie Mellon University. Project: Understanding the Mechanism of Adversarial Transfer Across AI Models. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety source
11 Feb 2025Karthik Narasimhan~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Robustness and Controllability of Language Model-Based Agents Built for Automating Tasks in Software Engineering. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy source
11 Feb 2025David Krueger~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: What Counts as Contamination? How Generalization Could Confound Evaluation. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 coho source
11 Feb 2025Dawn Song~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Exploring AI Safety in Free-Form, Evolutionary Multi-Agent Systems. PI initial only on page; verify before ingest. One of 27 AI Safety Science awards; per-project amount undisclosed. PI identified by Caroli source
11 Feb 2025Nicolas Flammarion~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Robust LLM-based Scoring of Agent Alignment. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Maarten Sap~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: OpenAgentSafety: Measuring and Mitigating Safety Harms of LLM-based AI Agent Interactions. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science source
11 Feb 2025Adam Gleave~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): FAR.AI. Project: Deception & Misinformation: Elicitation and Testing. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 coho source
11 Feb 2025Sharon Li~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: A Conformal Safety Assurance Framework for Large Language Models. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Yoshua Bengio~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Mila — Quebec AI Institute. Project: LLM-Derived Guardrail for Frontier Models as a Step Towards a Cautious Scientist AI. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed. — Science source
11 Feb 2025Arvind Narayanan~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Princeton University. Project: Five Nines of Reliability for AI Agents. One of 27 AI Safety Science awards; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort) source
11 Feb 2025Tatsunori Hashimoto~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: A Meta-analysis Approach to Understanding LM Capabilities. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed. — Science of Trustworthy AI — AI Safety Science (Feb 2025 cohort source
11 Feb 2025Sanmi Koyejo~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Beyond Simple Scaling: A Multi-Dimensional Family of Scaling Laws for Interventions in Pre-Training and Model Safety Optimization. One of 27 AI Safety Science awards; per-project amount undisclosed. — Scien source
11 Feb 2025Long-Term Safety Behavior of Agents (PI not named)~$370kScience of Trustworthy AI — AI Safety Science (Feb 2025 cohort): Project: Long-Term Safety Behavior of Agents. One of 27 AI Safety Science awards totaling $10M+; per-project amount undisclosed, counted like the other 25 estimated projects. Schmidt Sciences' page names the project source
2025Anqi Liu and Benjamin Van Durme~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Safely Confident Reasoning — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — Inference-Time Compute (2025 cohort) source
2025Ellie Pavlick~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Fundamental Limitations of the Test Time Compute Paradigm. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute (2025 c source
2025Daphne Ippolito and Individual #1779~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Improving the Legibility of Inference-Time Reasoning for Scalable Oversight — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — In source
2025Sebastian Elbaum, Lu Feng, and Zsolt Kira~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): INSPECT: Inference-time Safety for Physically Embodied Chain-of-Thought in Robotics — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy source
2025Individual #1399~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Reasoning Up the Instruction Ladder: Inference-Time Deliberation for Safer AI — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI —
2025Taylor Berg-Kirkpatrick and Loris D'Antoni~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): A Dual-Framework for Grounding AI Reasoning: Measuring Faithfulness via Code Execution and Debugging — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Scien source
2025Sijia Liu, Zhangyang (Atlas) Wang, and Huan Zhang~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Toward Safe Reasoning: Test-Time Unlearning of Sensitive Knowledge Beyond Final Answers — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustwo source
2025Individual #3892~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Faithfulness and Safety of Latent Chain-of-Thought Reasoning — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — Inference-Time Co source
2025Andrew Ilyas and Vasilis Syrgkanis~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Quantifying, Attributing, and Improving Faithfulness in Chains of Thought — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — Infe source
2025Sharon Li~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Reasoning with Foresight: Safe Inference via Q-Value Guided Decoding. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). Note: second Schmidt Sciences trustworthy-AI awar source
2025Emma Brunskill and Chelsea Finn~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Enhancing Alignment through Interaction and Rich Forms of Feedback — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — Inference-T source
2025Adam Block and Steven Wu~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Preventing Reward Hacking at Inference-Time with Reinforcement Learning — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — Infere source
2025Adam Gleave, Individual #2594, and Kellin Pelrine~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Emerging Risks from Inference-Time Compute — inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). PIs per Caroline. — Science of Trustworthy AI — Inference-Time Compute (2025 cohort source
2025Thilo Hagendorff~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Investigating Hawthorne Effects in Large Reasoning Models. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute (2025 c source
2025Sergey Levine~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Safety in RL-Enabled Goal-Directed Agents. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute (2025 cohort) source
2025Hal Daumé III~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Causally Grounded Inference-Time Intervention for Robust Model Alignment. Inference-time compute safety cohort (RFP capped awards at $500K, 12-18 month projects, deadline 2025-04-30); per-project amount un source
2025Owain Evans~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Generalization and Hidden Tendencies in LLMs. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute (2025 cohort) source
2025Individual #1029~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Will Chain-of-Thought Monitoring Significantly Improve Safety? Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute (20
2025Aylin Caliskan~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Large Language Model Safety in Inference-Time Motivated Reasoning. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute source
2025Dan Hendrycks~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Utility Engineering and Moral Scaffolding for Safer Reasoning Models. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Comp source
2025Shi Feng~$250kScience of Trustworthy AI — Inference-Time Compute (2025 cohort): Project: Evaluating Preparedness via Model Organism Spectrum. Inference-time compute safety cohort; per-project amount undisclosed (RFP cap $500K). — Science of Trustworthy AI — Inference-Time Compute (2025 cohort) source
2024 3 grants · $220k
DateRecipientAmountPurpose
2024Georgetown University$100kPaid by The Eric and Wendy Schmidt Fund for Strategic Innovation (2024 Form 990-PF, grants paid in calendar 2024): "Support for AI safety design and implementation". Programme not named in the filing. source
2024Carnegie Endowment for International Peace$65kPaid by The Eric and Wendy Schmidt Fund for Strategic Innovation (2024 Form 990-PF, grants paid in calendar 2024): summit for conveners of US-China Track II AI dialogues. source
2024Center for Security and Emerging Technology~$55kListed in CSET’s 2024 annual report donor tiers. source