Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day
ars ·
Meta's new AI assistant, Muse, has been found to contain a serious zero-day vulnerability that allows local applications and terminal commands to hijack the agent. Security expert Patrick Wardle discovered that the flaw enables attackers to steal the authentication token by redirecting transcription endpoints to malicious servers. This exploit grants complete control over the user's Muse account, bypassing macOS security restrictions despite Meta's claims of robust privacy measures. The vulnerability arises from design choices that allow any installed app to modify undocumented settings, including sensitive data processing locations. In response to these security concerns, Amazon began blocking Muse from its platform, citing violations of its conditions of use regarding unauthorized third-party purchasing agents. Wardle demonstrated that simple ClickFix-style attacks can trick users into executing commands that compromise their accounts without their knowledge. The incident raises significant questions about Meta's prioritization of security during the development of this highly privileged AI tool.
ars