Nearly a third of agentic AI pilot programs were halted over the past year, yet many remain active with their original credentials and production access intact. A Cequence Security survey of 202 enterprise leaders found that organizations are deploying these systems faster than they can control them.
Two-thirds of companies reported agents acting outside intended boundaries, with 29% experiencing measurable organizational damage. The disconnect between confidence and reality is stark: 94% believe their agents have appropriate access levels, but only 33% actually enforce least-privilege provisioning. Just 34% verify authorization at runtime, while 47% cannot reliably track which agents run in production and 46% lack basic 30-day audit trails.
The solution requires immediate action. Organizations should revoke credentials from retired pilots this week, implement runtime authorization checks, and inventory all agents before scaling further. Until decommissioned systems are properly logged out, enterprises remain vulnerable to uncontrolled autonomous actions.
Source: Cybersecurity Insiders, Independent CISO Research & Insights — https://www.cybersecurity-insiders.com/agentic-ai-pilots-live-credentials/