Advanced · Safety

Capability-Based Agent Authorization

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. Authorizing agents by handing them unforgeable, narrowly scoped capability tokens rather than checking roles at call time.

Technical Definition

Authorizing agents by handing them unforgeable, narrowly scoped capability tokens rather than checking roles at call time.

How it works

In a capability model, possession of a token is the authority to perform exactly the action it names, on exactly the resource it names, within its stated limits. This suits agents better than role-based access control: capabilities can be minted just in time, attenuated before being passed to a sub-agent, expired in minutes, and revoked individually. It also aligns with treating context as a security boundary — a prompt-injected instruction is harmless if the agent holds no capability for the action it demands.

Related Concepts

• Context as a Security Boundary — The design principle that an agent's context window is a trust boundary: what enters it defines what the agent can be persuaded to do.
• Agent Authorization — Deciding and enforcing what an authenticated agent is allowed to do, at what scale, and under which conditions.
• Agent Capabilities — The concrete set of actions an agent can perform — the tools, APIs, and side effects available to it.
• Agent Revocation — Withdrawing an agent's identity, credentials, or capabilities so it can no longer act — immediately and verifiably.