Research institute · Research orgs · Cambridge, Massachusetts
The Institute for Responsible Superintelligence (RESI) is a U.S.-based nonprofit research institute focused on “safe-by-design” foundations for advanced AI and eventual superintelligence. RESI’s central claim is that, as systems become too capable to rely on exhaustive post-deployment testing, safety must be expressed in advance as explicit properties and then achieved via mechanisms whose guarantees can be analyzed before deployment—drawing an analogy to how modern cryptography shifted from ad hoc patches to formal security definitions and provable constructions.
RESI positions itself as complementing (rather than replacing) other AI safety work such as evaluation, interpretability, alignment training, control, and deployment safeguards, by emphasizing formal definitions, analyzable guarantees, and their composition into larger systems involving models, tools, people, and institutions. Operationally, RESI describes an AI-assisted research approach (for example, harnesses to help theoretical alignment work throughout ideation, stress-testing, proof development, and experimentation) and an early-stage structure of working groups to pursue initial directions such as verification and delegation, modular architectures, incentives and strategic behavior, cryptographic mechanisms for AI safety, and legal mechanisms for superintelligence. RESI’s founding scientific leadership is drawn from theoretical computer science and cryptography, with roots in defining and proving security notions and impossibility results. RESI is fiscally sponsored by the Edward Charles Foundation (a 501(c)(3) public charity) while it builds its own nonprofit program activities through philanthropic grants (as publicly stated on RESI’s site).
No curated programs yet. Check the careers page ↗.
Publications are not yet indexed for this organization. Read them on resi.org ↗.
A Substack post announces a RESI initiative/workstream focused on ramping up formal-methods work and discusses how formal verification could support enforceable containment/guarantees in AI safety contexts.
RESI cofounder’s launch post (LessWrong)A LessWrong post from Adam Tauman Kalai describes RESI’s motivation and “safe-by-design” approach and states that he recently left OpenAI to cofound RESI with Shafi Goldwasser and Vinod Vaikuntanathan.
Boston Globe coverage on RESI’s approach and funding contextA Boston Globe article discusses RESI’s safe-by-design framing and provides context on leadership, the nonprofit structure, and philanthropic funding being cited as coming from Coefficient Giving and Schmidt Sciences, with the Edward Charles Foundation acting as fiscal sponsor (as summarized in the article).
MIT News clip referencing the RESI launch and mission framingAn MIT News ‘news clip’ references a Boston Globe story describing RESI and quoting the institute’s founders about developing mathematical principles/protocols for AI models to behave as intended.
RESI launches with a public agenda and founding announcement (introduced on X Aug 24, 2026; agenda published on resi.org)A research newsletter reports that RESI introduced itself on X on August 24, 2026 and published its agenda on resi.org, identifying Shafi Goldwasser, Vinod Vaikuntanathan, and Adam Tauman Kalai as founders.