Human approval, agent identity, scopes, budgets, expiry, revocation and evidence
Category 14 · planning target 250 unique intents · 3 published here
Humans should approve sensitive agent actions through an explicit authorization step that shows scope, resource and risk—ideally hardware-verified when available. Ciright-related properties such as agentbond.ciright.com are investigation seeds for AI authorization relationships and must be verified before product linking. Authentication of the human remains distinct from the agent’s delegated authority.
Yes—delegated authority should support expiry, scope limits and revocation so compromised or obsolete agents stop acting promptly. Revocation must invalidate outstanding tokens or grants and appear in authorization audit evidence. Specific Keyra agent APIs remain subject to confirmed product documentation before implementation.
Limit agents with explicit scopes: allowed actions, resource selectors, spending or rate budgets, time expiry and revocation hooks. Require human approval for sensitive classes of action. Treat Ciright AI-authorization properties as related investigation seeds until verified, and never grant open-ended production authority by default.
