Verification as a Control Primitive for Frontier AI

This independent, non-peer-reviewed technical report examines verification as a complementary control primitive for frontier AI safety. It develops a conceptual hazard model, a mechanism-neutral reference architecture, an evaluation agenda, and explicit limitations for connecting consequential action to independently verifiable authority. The report does not disclose proprietary implementation details or claim to solve alignment or eliminate catastrophic risk.

Author profile: https://niclasbraun.com
Research context: https://by4si.com