Anthropic expects artificial intelligence-based virtual workers to begin roaming corporate networks next year, said Jason Clinton, the company’s chief security officer.
Anthropic expects artificial intelligence-based virtual workers to begin roaming corporate networks next year, said Jason Clinton, the company’s chief security officer.
In an interview with Axios, he said that managing this AI identity data will require companies to reevaluate their cybersecurity strategies, or they risk exposing their networks to serious security breaches.
Agents typically focus on a specific, programmed task. In security, this means autonomous agents respond to phishing alerts and other threat indicators.
Virtual employees will take this automation a step further: AI identities will have their own «memory,» their own roles within the company, and even their own corporate accounts and passwords. They will have a level of autonomy that far exceeds that of today’s agents.
Anthropic sees itself as having two responsibilities: helping to address security concerns related to artificial intelligence. First, it’s rigorously testing Claude’s models to make sure they can withstand cyberattacks, Clinton said. Second, it’s tracking security issues and reducing the opportunities for attackers to abuse Claude.
Network administrators are already trying to track which accounts have access to different systems and fend off attackers who buy reused passwords for employee accounts on the dark web.
According to Clinton, while performing a task, AI employees could break the company’s continuous integration system, in which new code is merged and tested before deployment, and thereby violate the rules.
«In the old world, it’s punishable,» he said. «But in the new world, who’s responsible for an agent who’s been working for a couple of weeks and got to this point?»
Clinton says that the security of virtual employees is one of the biggest areas of security that AI companies can invest in over the next few years.
He is particularly interested in solutions that provide visibility into what an AI employee account is doing in the system, as well as tools that create a new account classification system that better accounts for virtual employees.
