Enterprises Deployed AI Agents Ahead of Governance Controls, Research Finds
Companies have deployed AI agents faster than the necessary controls to manage them are in place, a situation acknowledged by the organizations themselves. This is a central finding from five parallel surveys conducted by VentureBeat Research in June.
Enterprises have adopted AI agents at a pace that outstrips the development of necessary governance and control mechanisms, according to new research by VentureBeat. Organizations are aware of this gap and are now actively budgeting to implement standards and controls.
The research examined five key layers of agent governance: identity, evaluation, cost telemetry, context layer, and orchestration. Significant shifts are planned across these areas, with 57% to 68% of enterprises intending to switch or add vendors within the next 12 months.
Many deployed "agents" are, in fact, sophisticated chatbots rather than true autonomous systems. Only 10% of businesses reported that actual agents capable of multi-step tasks form the majority of their deployed systems, suggesting a widespread misunderstanding of the capabilities being utilized.
The research also highlighted that autonomy is outpacing trust in evaluation processes. Two-thirds of enterprises either allow agents to push changes to production based solely on automated evaluations, or are actively working toward this. However, only 5% fully trust these automated evaluations, and half of companies experienced customer-facing failures after agents passed internal assessments.
Furthermore, companies allowing agents to share credentials face a higher incidence of security issues. Sixty-nine percent of organizations permit credential sharing among some agents. These firms reported security incidents or near-misses at a rate of 63.5%, compared to 40.9% in companies where each agent has a distinct, scoped identity.