Least-privilege access

Least-privilege access (principle of least privilege)

AI

Least-privilege access means giving a person, app or AI agent only the access its job needs, so a mistake or misuse can reach as little as possible.

What is Least-privilege access?

Least privilege is an old security rule that matters more with AI agents, because an agent acts quickly and at scale. If a support agent can only look up orders and open returns, a confused or manipulated agent cannot delete products or export your customer list. How to apply it to agents: give each agent a narrow job, enable only the tools and data that job uses, assign it only the channels it should answer, and require approval for actions it decides on by itself, even within its access. Review access when the job changes. In Concierge, you tick only the capabilities each agent may use on its configuration page, and an agent without a capability cannot use it, however it is asked.

Example

For example, a store runs one agent for customer support with order lookup, returns and subscription changes, and a separate reporting agent with read-only access to sales data. Neither can do the other's job.

Why Least-privilege access Matters

Least privilege limits the damage from any error, misuse or prompt injection to the small set of things an agent was allowed to do.

Least-privilege access FAQ

Why give each AI agent separate access?

So no single agent can reach everything. If one makes a mistake or is manipulated, the harm stays within its narrow job.

Does least privilege slow agents down?

Rarely. AI agents with a focused job and fewer tools usually work faster and make fewer wrong choices.

Explore More Terms

Browse our complete glossary of e-commerce and subscription terms.

Space, at your fingertips

What would you do with 22% more revenue?