Safe Super Intelligence: Definition, Architecture, and Verification Standard

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Safe Super Intelligence (SSI) defines an architectural class of superintelligent systems where intelligence, capability, authority, safety, verification, governance, provenance, and consequential execution are explicitly separated and formally coordinated. The defining property of SSI is that the pathway from intelligence to externally consequential action is governed by machine-checkable conditions rather than just producing safe outputs. Under this architecture, an intelligence may observe, reason, learn, predict, plan, simulate, generate, refine, and propose without acquiring unrestricted authority to act upon the external world. A consequential action is executable only when it satisfies applicable constraints including type, capability, human authorization, governance, safety, verification, provenance, deployment-binding, temporal-validity, scope, and revocation. This standard establishes a canonical semantic and architectural layer encompassing Human-Guided Superintelligence (HGS), the Cognitive State Ledger (CSL), Russell Recursive Refinement (R^3), WAD-18 deterministic normative arithmetic, the SSI Axiomatic System, and the Safety Asset Class (SAC).