Sysdig launches runtime security and governance for AI agents

Sysdig has announced a governance-focused extension to its runtime security platform, delivering automated oversight for agentic AI through a product called Sysdig AI Defense. The solution is designed to observe the actions of AI agents in real time, from initiation to completion, and to block risky activities as they happen, with blocking responses intended to occur instantly and inline to prevent completion of the action.

The product builds a live inventory of agents, runtimes, models, MCP servers, skills and plugins using runtime telemetry, and it ties each agent’s actions to an accountable human. It enables detection of risky agentic actions, which can be held for human review or prevented before they finish, and offers predefined policy templates linked to frameworks such as the OWASP Top 10 for LLM Applications.

Each action is recorded as it happens, generating a live action graph that correlates what an agent claimed it did with what actually occurred, including visibility into kernel-level processes that might alter agent behaviour unseen at the user level.

SecurityWeek notes that the IBM Institute for Business Value previously warned about the rapid rise of autonomous AI agents and governance challenges as organisations move from pilots to production. Sysdig positions AI Defense as a purpose-built, automated runtime security and governance tool for AI agents, intended to provide auditable, enforcement-ready controls in real time.