AI Coworkers Will Break Current Access Security Models
The third wave of AI-driven work is on the doorstep: persistent AI coworkers that operate independently rather than within a single session or task. Security models built for chatbots and task-scoped agents are breaking down, as these digital colleagues need ongoing access instead of case-by-case approvals. Major vendors signal this direction, with Microsoft calling agents "digital colleagues" and OpenAI's Sam Altman outlining a virtual coworker vision back in early 2025.
Persistence fundamentally changes the risk profile. Today's AI agents typically rely on borrowed credentials through OAuth grants, in-session hand-offs, or general-purpose service accounts that place humans at the center. A true AI coworker holds access continuously, turning its credentials into a standing privilege much like a human employee's. Organizations must plan for front-loaded access or machine-friendly provisioning, along with the ability for AI coworkers to request additional permissions when tasks require them.
The industry remains unprepared: provisioning purpose-built identities for AI agents is still rare. Experts urge security teams to act now, addressing the full identity lifecycle for AI coworkers — including deprovisioning — before persistent agents become standard in the workplace and inherited credential models become a major attack surface.