[Schmidt Sciences](https://www.schmidtsciences.org)

May 17, 2026 (Full proposal deadline)

Tier 1: Up to $1M (1–3 years) | Tier 2: $1M–$5M+ (1–3 years); compute resources and engineering support available

Supports ambitious research advancing the scientific understanding, measurement, and control of risks from frontier AI systems.

Indirect costs capped at 10%; open internationally to non-profit institutions and universities.

--- ## Description Schmidt Sciences’ Science of Trustworthy AI program funds technical research to improve understanding, prediction, and control of risks from advanced AI systems, with the goal of enabling their trustworthy deployment. The program emphasizes deep, focused projects—especially ambitious, collaborative efforts—that can reshape the field’s understanding of AI risk. Supported research should align with a comprehensive research agenda spanning three main aims: (1) characterizing and forecasting misalignment in frontier AI systems; (2) developing generalizable measurements and interventions for AI evaluation and control; and (3) overseeing AI systems with superhuman capabilities, including addressing multi-agent risks. Successful applicants may also receive access to advanced computational resources, software engineering assistance, API credits with frontier model providers, and opportunities for community engagement. --- ## Due Dates - **Full proposal deadline:** May 17, 2026 (by 11:59 PM Anywhere on Earth) Notification of decisions is expected in Summer 2026. --- ## Funding Amount - **Tier 1:** Up to $1,000,000 total for projects lasting 1–3 years. - **Tier 2:** $1,000,000 to $5,000,000+ total for projects lasting 1–3 years. - Preference for Tier 2 is given to collaborative, multi-institution projects led by multiple PIs/labs. - **Additional support:** - Access to advanced computing resources (GPUs/CPUs, storage, networking) - Software engineering support via the [Virtual Institute for Scientific Software](https://www.schmidtsciences.org/viss/) - API credits with frontier model providers - Convenings and workshops Indirect costs are capped at 10% of the total project budget. --- ## Eligibility - Open to individual researchers, research teams, universities, national laboratories, institutes, and non-profit research organizations worldwide. - Lead applicant must be a non-profit or university (including international institutions with charitable status). - Multi-institution and cross-country collaborations are encouraged. - For-profit entities cannot lead projects but may participate as partners. - Indirect costs must not exceed 10%. - Projects must be charitable in nature and comply with relevant US and local laws. - No requirement for co-funding; awards can cover the entire project cost. --- ## Application Process - **Submission:** Applications must be submitted through the [Schmidt Sciences application portal](https://schmidtsciences.smapply.io/prog/science_of_trustworthy_ai_rfp_2026/). - **Materials required:** Detailed project proposal, budget (with indirect costs at or below 10%), and team information. - **Multiple proposals:** Applicants may submit more than one proposal per tier. - **Selection criteria:** - Research agenda fit - Scientific quality and rigor - Potential impact - Feasibility and scope - Team expertise and commitment - Cost effectiveness - **Non-competitive proposals:** Often lack focus, credible validity arguments for tools/evaluations, or concrete methods and outcomes. - **Award structure:** Typically provided as a series of one-year gifts or grants, with future years contingent on progress. - **International applicants:** May require equivalency determination for charitable status (Schmidt Sciences assists with this process). For detailed FAQ, see the [Trustworthy AI FAQ](https://docs.google.com/document/d/1nuqku7P9yQaxLPkeIhs10-21SEX77LwxexRwVmqSxkI/edit?usp=sharing). --- ## Additional Information ### Scientific Scope Projects should address one or more of the following: - **Characterizing and forecasting misalignment:** Understanding why advanced AI systems may act contrary to intended goals, especially under distribution shift or increased capability. - **Developing generalizable measurements and interventions:** Creating robust, decision-relevant evaluations and interventions that improve AI safety in novel or adversarial scenarios. - **Oversight for superhuman and multi-agent AI systems:** Methods for scalable oversight when direct human evaluation is infeasible, and addressing collective risks in multi-agent deployments. **Out of Scope:** Projects focused exclusively on interpretability (see [AI Interpretability program](https://www.schmidtsciences.org/ai-interpretability/)), jailbreak discovery, generic capability evaluations, near-term product engineering, fairness/bias research not tied to safety, policy/AI governance, content authenticity, or dangerous capability evaluations requiring security clearance. --- ## External Links - [Program page and application portal](https://schmidtsciences.smapply.io/prog/science_of_trustworthy_ai_rfp_2026/) - [Research Agenda](https://www.schmidtsciences.org/trustworthy-ai-research-agenda/) - [Trustworthy AI FAQ](https://docs.google.com/document/d/1nuqku7P9yQaxLPkeIhs10-21SEX77LwxexRwVmqSxkI/edit?usp=sharing) - [Program overview](https://www.schmidtsciences.org/trustworthy-ai/) - [Virtual Institute for Scientific Software](https://www.schmidtsciences.org/viss/) --- ## Contact Information | Contact | Details | |--------------------------|--------------------------------------------| | Program Email | trustworthyai@schmidtsciences.org | | FAQ | [Trustworthy AI FAQ](https://docs.google.com/document/d/1nuqku7P9yQaxLPkeIhs10-21SEX77LwxexRwVmqSxkI/edit?usp=sharing) | ---

2026 Science of Trustworthy AI RFP

Sign in to see full grant listing