Tenable details the agentic harness behind Hexa AI

Tenable has published details of the governance layer it built for Tenable Hexa AI, the agentic engine inside its Tenable One exposure management platform. The company calls this layer the harness, a runtime control environment that decides what context the model sees, which tools it can call, what must be validated before an action runs, when a human must approve, and what gets recorded afterwards.

Tenable said it treats its AI agents as untrusted insiders, granting them the entitlements of the person who invoked them and requiring explicit approval before any state change. Every action is traced and correlated by request identifier so it can be distinguished from human activity.

The vendor also runs two evaluation suites, one functional and one for safety, against every prompt, tool or model change, with more than a hundred curated cases scored against a pinned baseline.