1,000+ Opportunities
Find the right grant
Search federal, foundation, and corporate grants with AI — or browse by agency, topic, and state.
Schmidt Sciences has launched a pilot interpretability programme funding new methods for detecting and mitigating deceptive behaviors in large language models, with awards of 300,000 to 1,000,000 US dollars over one to three years.
The programme defines deception broadly and concretely: contradictions between what a model states and what it internally represents, misleading confidence claims, selective omission of information, sycophancy, manipulative discourse, and knowingly giving harmful advice. Three research directions are named. The first is identifying deceptive behaviors in language models, including monitoring and detection methods.
The second is creating targeted interventions and steering methods that improve model truthfulness. The third is implementing detection and steering techniques in practical settings, which the RFP illustrates with AI debate settings and decision support systems, along with building visualisations or dashboards that communicate model truthfulness to users and studying the role of deception mitigations in multi-agent interactions.
The call is open globally to individual researchers, teams, institutions and multi-institution collaborations, with a deadline of 26 May 2026 and informational sessions held on 2 April and 28 April 2026. It is distinct from Schmidt Sciences' broader Science of Trustworthy AI RFP and runs its own application portal.
Get a weekly digest of new grants like this
A free weekly digest of new foundation and federal funding opportunities as they're added to Granted. Unsubscribe anytime.
Or search similar grants →According to the current listing, eligibility includes: The programme welcomes individual researchers, research teams, research institutions, and multi-institution collaborations spanning universities, national laboratories, institutes and nonprofit research organisations. Applicants may be based anywhere globally, which makes this one of the more open interpretability funding routes and removes the national-eligibility barrier common to government AI safety programmes. Indirect costs must remain at or below 10 percent, a binding condition that requires institutional agreement in advance for most universities. Projects run one to three years with awards between 300,000 and 1,000,000 US dollars. Proposals should address at least one of the three named directions: detecting deceptive behaviors including monitoring methods, targeted steering interventions for model truthfulness, or practical implementation of detection and steering in applications such as AI debate settings, decision support systems, user-facing truthfulness dashboards, or multi-agent settings. Applications are submitted through the Schmidt Sciences portal at schmidtsciences.smapply.io under the 2026 interpretability RFP by 26 May 2026, with informational sessions held 2 April and 28 April 2026 at 1pm ET. Questions go to interpretability@schmidtsciences.org. This is a separate competition from the Science of Trustworthy AI RFP and a team may consider which is the better fit rather than assuming they are interchangeable. Confirm the full requirements in the official notice before applying.
The current listing shows awards range from 300,000 to 1,000,000 US dollars for projects lasting one to three years, so amount_min is 300,000 and amount_max is 1,000,000 as published. Indirect costs must remain at or below 10 percent, which is the figure most likely to catch university applicants: an institution with a 55 or 60 percent negotiated federal rate must obtain a waiver or absorb the difference, and that conversation with a research office should start before the proposal is written rather than after an award is offered. Because the ceiling covers up to three years, a 1,000,000 dollar award at full duration is roughly 333,000 dollars per year, which supports a small focused team rather than a large programme. Applicants should size duration and budget together rather than defaulting to the maximum on both. Verify award ceilings, matching requirements, and allowable costs in the official notice.
The published deadline was May 26, 2026, which has passed. Check the official notice for any future application windows before investing time in a proposal.
Schmidt Sciences 2026 Interpretability Request for Proposals for Detecting and Steering Deceptive Behaviors in Large Language Models is funded by Schmidt Sciences. Verify program details on the funder's official page before applying.
Start from the official opportunity page linked in this listing — it carries the sponsor's submission instructions.
Schmidt Sciences opened a 'Scaling AI Safety for a Multi-Agent World' program with awards up to $1 million and an August 8, 2026 deadline. Against a backdrop of federal research slowdowns, here is why private science philanthropy matters more than ever, how these funders differ from federal agencies, and how researchers should approach them.
Read articleSchmidt Sciences' 2026 Science of Trustworthy AI RFP closes May 17 with two funding tiers — up to $1M (Tier 1) and $1–5M+ (Tier 2) over 1–3 years, with a 10% indirect cost cap. The three research aims target misalignment under distribution shift, predictive-validity evaluations, and oversight of superhuman systems. Here is why the structure favors team-based proposals.
Read articleThe U.S. Government Policy for Stopping High-Risk Life Sciences Research, released July 28, 2026 under EO 14292 and implemented at NIH through NOT-OD-26-101, replaces mitigation with prohibition. It bans dangerous gain-of-function research outright, gates potential DGOF behind government-wide review, and creates International Research of Concern — a research-security regime wearing biosafety clothing. Penalties run to five-year debarment and False Claims Act exposure.
Read article