Anthropic's Chief Information Security Officer Jason Clinton predicted that virtual AI employees could begin operating within corporate networks by next year. These AI agents would possess unique corporate accounts, passwords, and 'memories'—a leap from current task-specific AI agents.
Clinton highlighted the urgent need for companies to develop cybersecurity protocols to govern access, monitor privileges, and determine accountability for AI-driven decisions. He believes these virtual employees will represent the next AI frontier, posing major security risks if their identities and actions aren't properly tracked.
Anthropic is already focused on fortifying its Claude models to endure cyberattacks and minimize misuse. Clinton expressed concern that virtual employees might operate autonomously, causing complications such as unauthorized code deployments or acting as rogue agents within systems.
With companies like OpenAI and Anthropic investing heavily in AI infrastructure, Clinton emphasized that better visibility and role classification systems are critical. He also noted the market’s shift toward managing non-human identities, pointing to Okta’s newly launched unified control platform as one example.
This Axios interview outlines how AI workers will reshape workplace security, urging firms to revisit how they manage digital identities, credentials, and autonomous systems.