Delinea CEO Art Gilliland says enterprises deploying autonomous AI agents across critical systems face new identity risks, even when those agents have legitimate access. He argues that continuous monitoring and runtime authorization are needed to detect harmful activity and control agent access. Gilliland describes access as “oxygen for AI,” emphasizing that AI must be able to reach data to function. He recommends distinguishing human privileges from those granted to AI agents and adopting intelligent authorization to monitor behavior, spot credential misuse and make contextual access decisions in real time. In an ISMG interview he notes that many firms focus on endpoint and network defenses while leaving legacy identity weaknesses and standing privileges unaddressed, and that Delinea’s use of frontier AI models has uncovered roughly six times more vulnerabilities year over year. 1