The Nescience Manifesto: A Proposal for a Global, Ownerless Institution to Audit Frontier AI

Nescience is a proposal for a global, non-profit, ownerless institution whose single function is to detect when a frontier artificial-intelligence system stops being trustworthy, and to hold binding power to pause or withhold its deployment. The manifesto argues that current AI safety practice faces a verification gap: interpretability understands only a small fraction of frontier-model computation, and measured capability shows little correlation with measured safety. It proposes three artifacts — a "police model" of independent lineage trained solely for detection and deliberately less capable than the systems it monitors; a forensic catalog of behavioral and activation-level signatures of deception, sabotage and alignment faking; and a control infrastructure combining trusted monitoring, resampling, distillation, consistency interrogation and non-model hard limits. The institutional design addresses the "who watches the watcher" problem through an irrevocable endowment with no majority bloc, an eleven-seat board on which AI labs hold no vote, an external oversight commission with full access, a six-year non-renewable directorship with post-tenure quarantine, a capped citizen fund that elects the civil-society bench, and a "death clause" under which dissolution cryptographically destroys the institution's sealed assets rather than transferring them. Published in Portuguese (original) and English. Conceived by Lucas Parenti, Porto Alegre, Brazil, on October 19, 2026, and written in collaboration with Claude (Anthropic). Released under CC BY 4.0 so that any party may adopt, adapt or found the proposal, with attribution.