Cloudflare Introduces Agent Access Model for AI Security

Cloudflare announced the Agent Access Model (AAM), a security reference architecture designed to secure task-scoped AI agent workloads. Relying on prompts to enforce security boundaries is inherently flawed as agents can be manipulated and operate at machine speed. The Agent Access Model shifts authorization outside the model, enforcing real-time controls and continuous mediation scoped directly to specific tasks using short-lived credentials.