The Agent Access Model

Cloudflare is proposing a new security framework for AI agents. Traditional models assume humans—people working at human speed, leaving audit trails. Agents are different. They operate at machine speed, potentially hitting dozens of systems in seconds, and their behavior can be manipulated by injected data. Conventional security rules fail quietly at that speed. A prompt instruction like 'don't access production' can't actually enforce a boundary. Cloudflare's Agent Access Model takes a different approach. Every action an agent takes gets authorized in real time against what that specific task is allowed to do. Agents get short-lived credentials that expire when the task ends, enforcement happens at the network and tool layer, and capabilities can only shrink as a task runs—never expand. It's a rethinking of access control built for software that doesn't work like humans.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton