Anthropic says virtual AI workers will be here within a year

American artificial intelligence startup Anthropic expects AI-powered virtual employees to start appearing in corporate networks next year, the company's chief information security officer Jason Clinton said in an interview with Axios.

As Clinton noted, companies will have to review their cybersecurity strategies for managing virtual workers, as it could lead to serious security breaches in their networks.

Despite the potential risks, Clinton says virtual workers will be the next big innovation in AI. Today’s agents typically perform specific tasks, such as detecting fraudulent alerts or cybersecurity threat indicators. Virtual workers will extend that automation by giving them “memory,” defined roles within the company, and even their own accounts.

The main challenges Anthropic identifies are defining permissions for virtual workers, securing their work accounts, and monitoring their activities. The company believes it has two key obligations to such workers: first, to thoroughly test Claude models to ensure they are resilient to cyberattacks, and second, to monitor potential security issues and prevent abuse by attackers.

Clinton notes that network administrators are already working to control access to various systems, protecting them from attackers using resold passwords from the dark web. He also emphasizes that virtual workers could become a new problem that could break companies' continuous integration systems, where new code is merged and tested before a stable release.

The cybersecurity chief at Anthropic says that securing virtual workers will be a top priority for AI companies, and they will be investing heavily in it over the next few years. Clinton is particularly interested in solutions that will track the activity of virtual worker accounts, as well as tools to create a new account classification system that better addresses the needs of AI workers.