Confidence in AI Agents Don't Match Controls
Harness, Thursday, September 10th, 2026
A survey of 700 organizations finds confidence in AI agents outrunning the controls to test, secure and roll them back.
A Harness survey of 700 organizations documents a gap between stated confidence in AI agents and the controls needed to test, secure, track and roll them back.
The rollback finding is the most striking, since the ability to reverse an agent's actions is a basic operational requirement that many organizations have not established, meaning a misbehaving agent's changes must be unpicked manually.
The survey data is useful for security and platform leaders making the case that agent governance should precede rather than follow wider deployment.