Runtime AI security, also called runtime security for AI agents, is the enforcement of security policy on a model or agent while it is in use. Each prompt, response and tool call is checked in the request path before it proceeds.
How it differs from posture management
Posture management inventories models, agents and configurations and reports what is exposed. It works before and between runs. Runtime security works during the run, on the call that is happening now.
How it differs from detection
Detection observes an action and raises an alert. Runtime security decides before the action executes. For an agent with permission to move money or change records, that is the difference between a report and a prevented loss.
What it has to cover
The prompt, the response, the tool call and the record of each decision. A control that covers the first two is a set of AI guardrails. A control that covers all four, and ties each action to the person whose authority the agent used, is runtime governance.
The full treatment is on the runtime AI security page.
Further reading: Runtime security and runtime governance · The CISO's guide to AI security categories.
Related terms
Runtime governance · AI agent security · AI guardrails · MCP security · Prompt injection · 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