VentureBeat Research has released findings from five parallel surveys conducted in June 2026, highlighting significant gaps in governance for enterprise AI agents. Enterprises have frequently deployed AI agents without the necessary controls in place, resulting in a need to retrofit systems to meet internal standards.
The research outlines five critical control layers: identity, evaluation, cost telemetry, the context layer, and orchestration. Identity controls which agent can perform specific tasks under certain credentials, while evaluation assesses the quality of the agent's work. Cost telemetry tracks operational expenses, the context layer provides necessary business data, and orchestration coordinates multi-step tasks. Despite these needs, many organizations have not fully implemented these controls, with only 10% of enterprises running true multi-step agents.
The surveys revealed that two-thirds of enterprises are either allowing agents to make system changes based on automated evaluations or are planning to do so. However, only 5% fully trust these evaluations. Security concerns were also noted, as 69% of companies allow agents to share credentials, leading to a higher incidence of security incidents.
The research further indicates that enterprises are not fully utilizing their AI hardware, with over 80% reporting GPU utilization at 50% or less. Additionally, governance issues in business context data have led to incorrect agent responses, with 57% of enterprises experiencing such issues.
The target audience for this research includes enterprise decision-makers and stakeholders responsible for AI deployments. The insights are crucial for those looking to enhance AI agent governance and improve system reliability and security.
Work implications: These findings highlight the need for roles focused on AI governance and security, potentially augmenting roles in IT management and data governance.
Originally reported by VentureBeat