Issue IT governance models have no account of systems that decide and act on their own

Summary

Deloitte identifies governance and control as the third infrastructure obstacle to agentic AI. Enterprises find it hard to set up suitable oversight for systems that are designed to operate on their own. Conventional IT governance assumes a human makes the decisions, so it does not cover AI systems that decide and act independently. The report argues the difficulty goes beyond technical controls to basic questions about process design, because many organisations try to automate current processes instead of reimagining workflows for agents. Its later sections sketch responses such as graduated autonomy levels, human supervisors at designed handoff points and zero trust authentication for agent actions. Without such mechanisms, it suggests, agents cannot safely be given greater independence.