A structural gap in OAuth 2.1 and JWT standards means that AI agents authenticated through OAuth carry no standardized identity for the agent instance, the task being executed, or the authorizing human. This makes it impossible for SIEM, IGA, and PAM controls to distinguish legitimate AI agent activity from compromised or hijacked agent behavior, leaving organizations with broad OAuth-scoped AI agent deployments — including MCP-connected Claude Code and GitHub integrations — unable to detect or investigate agent-sourced access abuse.