Delegated authority (agent delegation) is the mechanism by which a user grants an AI agent permission to act on their behalf, ideally scoped narrowly to a specific task, limited in time, and revocable. It is how you let an agent do useful things without handing it the keys to everything.

The danger with autonomous agents is an agent with broad, standing access to a user’s accounts. Delegated authority is the discipline that prevents that.

How it works

Rather than giving an agent a user’s full credentials or open-ended access, delegated authority issues the agent tightly scoped permission: do exactly this task, up to these limits, for this long, then the authority expires. This is least privilege and just-in-time authorization applied to agents. If the agent is hijacked or misbehaves, the damage is bounded by its narrow grant, and the authority can be revoked without disrupting the user’s own access.

Why it matters

Delegated authority is foundational to safe agentic commerce and agent-based workflows. It turns "an agent acting for me" from an all-or-nothing risk into a controlled grant: an agent allowed only to book travel under a set amount cannot drain an account, and one allowed only to read cannot transact. As agents proliferate, ephemeral, scoped delegation, paired with agent identity and fine-grained authorization, is what keeps their power in check.

Frequently asked questions

What is delegated authority for AI agents?

Granting an agent scoped, time-limited, revocable permission to act on a user’s behalf, rather than full access.

Why does it matter?

It bounds the damage if an agent is hijacked or misbehaves, and lets authority be revoked without disrupting the user.

How does it relate to least privilege?

It’s least privilege and just-in-time authorization applied to AI agents.

Related: Fine-Grained Authorization (FGA) · Principle of Least Privilege · Agent Identity · AI Agents / Agentic AI · Agentic Commerce