Control effectiveness for AI agents
Design and operating effectiveness stop short for AI agents. Control effectiveness asks whether a control held for every action, and whether it could be better.
Insights
Insights
Perspectives on AI Accountability: proving what AI agents actually did.
The Prove it series
Six perspectives on proving what AI agents actually did. One a week.
Seeing isn't proving
Runtime visibility shows what an AI agent did. Evidence proves who acted and whether the control held, to people who were not in the room.
“We think we know” isn’t an answer
A green dashboard is not proof. An honest board report says held, did not hold, or insufficient evidence for every AI control.
Behaviour change, not just blocking
Publishes 20 October 2026
Evidence you don't have to take our word for
Publishes 27 October 2026
Controls that get better
Publishes 3 November 2026