Enterprise AI Organizations Struggle with Deployment, Trust, Security, and Cost Optimization

A Scramble for Control as AI Agents Outpace Governance

Enterprises deploying AI agents have outpaced their ability to govern them, according to a series of parallel surveys conducted by VentureBeat Research in June 2026. The findings reveal that 57 to 68% of companies plan to switch or add vendors for AI agent controls within 12 months, signaling a scramble to address fundamental gaps in trust, security, and operational oversight.

Five Critical Control Layers Under Scrutiny

The research focused on five critical control layers required to manage AI agents: identity, evaluation, cost telemetry, context, and orchestration. These controls govern who can act, how decisions are validated, financial impacts, data reliability, and workflow coordination. However, most enterprises lack deployment clarity. Seventy-one percent of respondents reported that fewer than a quarter of their “agents” can handle multi-step tasks independently, with only 10% claiming true agents dominate their systems.

Automated Evaluations Without Human Oversight Pose Risks

A major risk lies in the reliance on automated evaluations without human oversight. Two-thirds of enterprises allow agents to push code or system changes to production based solely on automated assessments, despite only 5% fully trusting these evaluations. This practice has already led to failures: half of companies experienced a customer-facing issue from an agent that passed internal evaluations. The report warns that enterprises should test evaluations against real-world outcomes rather than internal benchmarks before removing human review.

Security Lapses Widespread as Credential Sharing Rises

Security vulnerabilities are also widespread. Sixty-nine percent of companies permit agents to share credentials, such as using a single API key across multiple systems. Organizations allowing credential sharing faced security incidents or near-misses at a 63.5% rate, compared to 40.9% for those with scoped identities for each agent. The study emphasizes that scoped identities—where each agent has unique, limited access—are critical for preventing breaches, particularly for agents interacting with production systems.

Cost Management Challenges Plague AI Infrastructure

Cost management remains another challenge. More than 80% of enterprises running their own GPUs report utilization rates of 50% or lower, with only 44% rigorously tracking AI compute costs and returns. The research advises prioritizing optimization of existing infrastructure over acquiring more hardware, highlighting that underutilized resources represent a significant financial inefficiency.

Data Context Gaps Threaten Agent Reliability

Governing the data agents use is equally urgent. Fifty-seven percent of enterprises traced confident but incorrect agent responses to unmanaged or inconsistent business context, such as outdated metrics or missing documents. This highlights a foundational issue: without centralized governance of definitions and data, agents cannot reliably inform decisions. The report stresses that addressing context layers must precede scaling agent deployments.

Market for AI Controls in Turmoil as Enterprises Hunt Alternatives

The market for AI agent controls is in flux. No single vendor dominates any of the five control layers, with enterprises actively exploring alternatives. Orchestration, in particular, sees the highest switching intent, with 68% planning to adopt, add, or replace platforms within 12 months. However, the direction of this shift—whether toward built-in platform tools or specialized solutions—remains unclear, setting the stage for significant market evolution over the next year.

Survey Methodology Reveals Systemic Gaps in AI Governance

VentureBeat Research surveyed 573 professionals at organizations with 100+ employees, with each report detailing methodology and findings. The data collectively underscores a systemic gap: enterprises are deploying AI agents at scale without the infrastructure to ensure their reliability, security, or efficiency. As organizations rush to retrofit controls, the report serves as a cautionary guide for balancing innovation with accountability.