AI Agents Outpace Company Controls: A Known Risk

Enterprises have knowingly deployed artificial intelligence agents ahead of establishing the necessary management controls, according to five parallel surveys conducted in June 2026. A significant majority of these deployed 'agents' are in fact basic chatbots, with only a small percentage being true multi-step AI. This deliberate deployment has led to a current focus on retrofitting and budgeting for new controls.

Respondents indicated plans to switch or add vendors for agent controls within the next 12 months. Key control areas include identity management, evaluation of agent performance, cost tracking, data context, and orchestration. Many companies have experienced issues, such as AI agents causing customer-facing failures after passing internal evaluations. Security incidents are also more frequent in organizations that allow agents to share credentials.

Furthermore, AI compute resources, including GPUs, are often underutilized, with more than 80% of enterprises reporting 50% or less capacity usage. A significant portion of companies struggle to rigorously track AI compute costs. Inaccurate agent answers have also been traced to missing or inconsistent business context, highlighting the need for governed data definitions before scaling AI agents.