Meta Patches Muse Exploit Allowing AI Agent Control

Meta has released a critical patch for its Muse macOS app following the discovery of a significant zero-day vulnerability. This flaw could have allowed attackers to gain control over the AI agent within the application. The vulnerability, identified by security researcher Patrick Wardle, exploited an undocumented setting within Muse. This setting enabled potential attackers, who had local access to the user's device and were running local code, to redirect the app's transcription processing. Instead of sending data to Meta's secure servers, the transcription could be rerouted to an attacker's own endpoint. This redirection would grant the attacker unauthorized access to the user's Muse account, as reported by Ars Technica. Several design decisions within the Muse application reportedly contributed to the existence and severity of this flaw. These included the choice to process Muse dictation in the cloud rather than directly on the device, which created an opportunity for interception. Additionally, the design allowed any application to control all of Muse's undocumented settings, providing a broad attack surface for malicious actors. While the exploit required local access, its potential to compromise AI agent control and user accounts made it a serious concern. Meta's swift action in patching the vulnerability addresses these critical security risks.