AI agent governance: why 74% roll back

A Sinch survey of more than 2,500 enterprise leaders, run in early 2026, landed a number worth sitting with: 74 percent rolled back a live AI customer-service agent. Stranger still, the rollback rate climbed to 81 percent among the teams with the most mature governance. That is not a failure of governance. That is governance working.

What AI agent governance actually caught

The mature teams rolled back more because they could see more. They had the logging, the review, and the alerting to catch an agent doing confident, wrong things, and the discipline to pull it. The teams without governance are not safer. They just cannot see what their agent is doing. Tellingly, 98 percent still plan to grow AI spend, and 76 percent are redirecting it toward trust, security, and control.

What it means for a small business

You do not need a governance department. You need the same few things on any agent before it touches a customer: a tight scope, a human checkpoint on anything irreversible, full logging, and an alert when it is unsure. That is the difference between an agent you can leave running and a 2am apology.

• Scope every agent to the one system it needs, and nothing more.
• Put a human in front of anything irreversible: money, customer messages, deletions.
• Log every decision and alert the moment it is unsure, so silence never means 'fine.'

Bring me the agent you're tempted to leave running, and I'll tell you what I'd put around it before it touches a customer.