Enterprises Knowingly Deployed AI Agents Without Adequate Governance, VentureBeat Research Finds

Enterprises have knowingly deployed AI agents without the necessary governance controls in place, according to five parallel surveys conducted by VentureBeat Research in June 2026. A significant majority of these organizations, ranging from 57% to 68% across five measured control layers, plan to switch or add new vendors within the next 12 months to address these gaps. Approximately one-third of companies are looking to make these changes within the next quarter. The research identified five critical control layers for trusting AI agents: identity, evaluation, cost telemetry, context layer, and orchestration. Most deployed "agents" are actually simple chatbots, with 71% of enterprises reporting that a quarter or fewer of their agents can perform multi-step tasks autonomously. Only 10% of companies stated that true agents constitute the majority of their AI deployments. Furthermore, two-thirds of enterprises either currently permit agents to push code or system changes to production based solely on automated evaluations or are planning to do so within a year, despite only 5% fully trusting these evaluations. Half of companies have experienced customer-facing failures from agents that passed internal evaluations in the past year. Security is also a concern, as 69% of companies allow agents to share credentials, leading to higher rates of security incidents or near-misses compared to organizations with scoped identities for each agent. In terms of infrastructure, over 80% of enterprises running their own GPUs report utilization rates of 50% or less, with only 44% rigorously tracking AI compute costs and returns. Finally, 57% of enterprises have encountered confident but incorrect agent answers in the last six months, often stemming from missing or inconsistent business context data, highlighting the need for governing the data agents use.