Enterprises Struggle With AI Agent Security and Trust Issues

WHO:

VentureBeat - 107 enterprises participated in the survey regarding AI agent security.

VentureBeat - 157 enterprises participated in the survey regarding AI agent evaluation.

WHAT:

VentureBeat - 71% of enterprises report that a quarter or fewer of their deployed "agents" are true multi-step orchestrated workflows rather than single-prompt chatbot wrappers.

VentureBeat - 51% of enterprises expect a hybrid control plane for agent orchestration by the end of 2026.

VentureBeat - 27% of enterprises have no real-time way to stop a runaway agent before incurring costs.

VentureBeat - 58% of enterprises either run a governed semantic layer in production (25%) or are building one (34%).

VentureBeat - 57% of enterprises plan to switch or add a retrieval provider within twelve months.

VentureBeat - 54% of enterprises reported having experienced a confirmed AI agent security incident or a near-miss.

VentureBeat - 18% of organizations reported a confirmed incident, while 36% reported a near-miss.

VentureBeat - Only 32% of enterprises give every AI agent its own scoped, managed identity.

VentureBeat - 30% of enterprises isolate their highest-risk agents in sandboxes.

VentureBeat - 66% of organizations allow or are engineering toward zero-human-in-the-loop deployment for low-risk agents.

VentureBeat - 50% of organizations have shipped an AI feature that passed internal evaluations but failed in production.

VentureBeat - Only 5% of organizations fully trust automated evaluations.

WHERE:

VentureBeat - The survey included enterprises from various industries, including Technology/Software, Manufacturing, Retail/E-commerce, and Healthcare/Life Sciences.

WHEN:

VentureBeat - The survey was conducted in June 2026.

WHY:

VentureBeat - Enterprises are experiencing a security gap due to the rapid proliferation of AI agents outpacing the development of necessary security controls.

VentureBeat - The evaluation gap exists because organizations are granting more autonomy to AI agents while trusting the evaluations less.

HOW:

VentureBeat - Enterprises primarily use provider-native security tools from OpenAI, Google, and Microsoft for AI agent security.

VentureBeat - The evaluation stack is fragmented, with many organizations relying on provider-native evaluations or having no dedicated evaluation tooling at all.

A recent report from VentureBeat reveals that 71% of enterprises indicate that only a quarter or fewer of their deployed agents operate as true multi-step orchestrated workflows, with many still relying on single-prompt chatbot wrappers. Additionally, 51% of businesses anticipate implementing a hybrid control plane for agent orchestration by the end of 2026, while 57% are looking to switch or add a retrieval provider within the next year. Furthermore, 58% of enterprises are either running a governed semantic layer in production or are in the process of building one, though 27% still lack a real-time method to halt a runaway agent before incurring costs.