Establish a standardized agent sandbox attestation process where autonomous AI agents must pass formal verification tests — adversarial prompts, scope-boundary tests, and rollback audits — before being authorized to act in public-facing contexts. Attestation certificates expire and must be renewed after major model updates.