Institute for Responsible Superintelligence

RESI frames this as a foundational science problem analogous to how modern cryptography progressed: researchers formulated precise security definitions and then built primitives whose guarantees were analyzable under explicit assumptions. RESI’s mission is to build scientific foundations that help make superintelligence safe by design, using cryptography and related formal methods, while also highlighting additional relevant domains (e.g., incentives/economics and legal mechanisms) for designing robust systems and governance. RESI’s planned outputs include (1) mapping which safety properties are meaningful, achievable, or impossible under stated assumptions; (2) developing mechanisms, protocols, and architectures that provide useful guarantees and studying how these guarantees behave under composition into larger systems (models, tools, people, institutions); and (3) building implementations and proofs of concept of promising constructions, with the expectation that some items could be relatively low-cost while others might require frontier-model training and substantial resources. RESI also describes “harnesses” intended to make its internal research more AI-assisted across tasks such as ideation, stress-testing definitions, and checking proofs, and it plans working groups that will open new directions and revisit classical questions through the lens of superintelligence safety.