Despite executives expressing strong confidence in their AI governance, a recent survey reveals that most organizations grant AI agents broad, often unchecked, access to sensitive systems. This misalignment creates significant risk, with nearly a third of respondents reporting tangible harm from unrestricted AI deployments. The research also found that 42% of breaches involving AI were attributed to over‑permissive access, highlighting the tangible cost of complacency.
The study highlights a critical gap between security and controls. While 78% of firms claim robust access management, only 31% enforce least‑privilege policies, allowing AI models to read, write, or modify data without sufficient oversight. Moreover, 25% of surveyed companies admitted they lack AI‑access policies altogether, relying on rules that are hard to audit.
The findings underscore the urgent need for tighter AI access frameworks. Companies must move beyond confidence assumptions, implement permissions, and monitor AI behavior to prevent costly breaches and protect user trust.
Source: Read original article