Three tiers of enforcement for AI agents - strong, bounded, detectable

Most AI agent frameworks give you zero enforcement. Your agent can call any tool, take any action,...