Ajar

Ajar: Measuring Open Privilege in Agent Defenses An agent-security benchmark reports two numbers, attack success and benign utility, and both are read off runs that happened. Neither says what the defense stood ready to allow on the paths no run took. Ajar asks it directly: for each benign task i...

Ajar: Measuring Open Privilege in Agent Defenses

A language model agent acts through the tools it is given. The data it reads while working on a task can redirect what it does with those tools. A growing set of techniques for safe and secure agent execution therefore sits between the agent and its tools, aiming to enforce access control,...