Enterprises are deploying AI agents faster than they can verify those agents are safe to run, according to new research published in July 2026. A VentureBeat Pulse survey of 157 organizations found that half have shipped an agent or LLM feature that passed internal evaluations and then caused a customer-facing failure — and only 5% fully trust automated evaluation today.
The Evaluation Gap: Autonomy Outpacing Trust
Enterprises are granting AI agents more autonomy while trusting the tests meant to govern that autonomy less. According to VentureBeat’s Pulse Research, the most-cited weakness in current evaluation practices is that tests “align poorly with real-world outcomes,
Related news
- Agentic AI Is Untamable: Ask the Right Security Questions – Dark Reading
- Google DeepMind boss Demis Hassabis issues call to action on AI safety standards – IT Pro – Google News – AGI
- Enterprise AI Will Be Defined By Trust – Forbes Tech
Sources
- The Three Dimensions of Custom Agentic Alignment: Purpose, Principles and Practices – Towards Data Science
- The agent evaluation gap: Enterprise AI organizations have a reality-alignment problem, not a coverage problem — and most are shipping to production anyway – VentureBeat
- ‘No company is going to go to jail for you’: Proton’s CTO on balancing privacy, policy, and trust – The Verge
- A Stalled AI Pilot Isn’t A Technology Failure. It’s An Alignment Failure. – Forbes Tech
- Industry Reactions to Pentagon Suspending CMMC Phase 2: Feedback Friday – SecurityWeek
