AI Agent Security

A chat model produces text. An agent takes actions. The security question moves from what was said to what was done, and on whose authority.

AI agent security is the set of controls on software that uses a model to take actions: calling tools, changing records, sending messages, moving money. It covers what the agent can reach, what it is allowed to do there, and whose authority it is acting on.

Why agents change the problem

A chat application produces text and a person decides what to do with it. An agent acts on its own output. A manipulated prompt that once produced a wrong answer now produces a wrong action, at machine speed, under a credential borrowed from a person.

What it covers

Reach. Which tools and servers the agent can call. See MCP security.

Authority. What the agent is permitted to do, and on whose behalf. See agent authority scope.

Enforcement. A check on each action before it runs. See runtime AI security.

Approval. A named person deciding the actions that fall outside the agent's authority. See human in the loop.

Evidence. A record of what the agent did and who decided. See tamper-evident audit log.

Further reading: Runtime security and runtime governance · The CISO's guide to AI security categories.

Runtime AI security · Runtime governance · Agentic AI · Agent authority scope · MCP security · Full glossary

Put this in the path of your own agents.

Policy enforced inline between your agents and every model and tool they reach, with a record bound to the human who owns it.

Request a Demo Read the docs