AI Agents as Security Principals: Why Enterprise Identity Stacks Are Failing

As enterprises deploy AI agents without adequate security oversight, a critical vulnerability emerges: identity systems designed for human users with bounded sessions cannot contain autonomous agents operating indefinite loops with persistent credentials. This week's announcements from GitHub, Microsoft, and Cloudflare reveal that the industry finally recognizes the scope of this agent security crisis and proposes three competing architectures—signaling that fundamental changes to enterprise access control are inevitable.