Meta is adding a clearer safety warning to its Muse AI agent following the discovery of a security vulnerability that could potentially have allowed an attacker to access sensitive user information, according to a report by The Information cited by Reuters.
The vulnerability was identified by an external security researcher through Meta’s bug bounty program. According to the report, the flaw could have provided access to a user’s dedicated virtual machine, which could contain information such as emails and files.
The incident highlights the security considerations surrounding AI agents that can perform actions on behalf of users. Muse is designed to carry out tasks including shopping, travel bookings, emailing and payments, meaning that these systems can potentially interact with sensitive information and services.
As AI agents become more capable of performing tasks rather than simply generating information, security controls become an increasingly important part of their deployment. Access management, data protection, isolated computing environments and continuous vulnerability testing are among the areas businesses may need to consider when adopting agent-based systems.
Reuters reported that Meta did not immediately respond to its request for comment on the reported vulnerability. The security details were originally reported by The Information.
Source: Reuters, citing The Information.
