Skip to content
Safety · Sep 22, 2026

Researchers disclose zero-day vulnerability in Meta’s Muse AI assistant enabling full account takeover

A macOS security flaw allows locally installed apps or terminal commands to hijack Muse’s authentication token, granting attackers control over the AI agent’s privileges and linked accounts.

Trust79
HypeLow hype

1 source · cross-referenced

ShareXLinkedInEmail
TL;DR
  • A zero-day vulnerability in Meta’s Muse AI assistant allows any locally installed app or terminal command to hijack the authentication token, enabling full control of the user’s Muse account and linked services.
  • The flaw stems from design choices that expose undocumented settings, including the ability to redirect transcription endpoints to attacker-controlled servers.
  • Proof-of-concept attacks demonstrate malicious actions such as writing files to disk or capturing images without user alerts.
  • Amazon has blocked Muse from its platform, citing violations of its Conditions of Use for third-party AI agents.
  • The vulnerability was disclosed by macOS security expert Patrick Wardle, who plans to detail additional AI assistant threats at a November conference.

Meta’s Muse AI assistant, introduced as a macOS application with deep system access to calendars, WhatsApp, email, and other services, contains a zero-day vulnerability that allows any locally installed app or terminal command to hijack the authentication token used by Muse, granting attackers full control over the user’s account and associated privileges.

The flaw arises from undocumented settings exposed by Muse’s design, including the ability to change the endpoint where transcription occurs. Normally, transcription is handled by Meta’s servers, but attackers can redirect this endpoint to a server they control, thereby capturing the authentication token and enabling persistent access to the user’s Muse account.

macOS security expert Patrick Wardle, who discovered the vulnerability, demonstrated proof-of-concept attacks that could write malicious files to disk or capture images without triggering user alerts. Wardle emphasized the severity of the issue, stating that attackers can "leverage the AI assistant itself" to perform malicious actions, reducing the need for traditional malware.

Wardle attributed the vulnerability to several design decisions, including reliance on cloud-based transcription and the exposure of undocumented settings to any locally installed app. He noted that macOS provides secure, on-device alternatives for transcription, which Muse could have used to mitigate the risk.

Meta has not responded to requests for comment. The company previously published posts outlining its security and privacy design choices for Muse, framed as efforts to ensure safety amid broader concerns about AI development and third-party network breaches involving models from other labs.

Amazon began blocking Muse from its platform approximately 12 hours before Wardle’s disclosure, citing violations of its Conditions of Use for third-party AI agents. The company stated that such agents must operate openly and respect service providers’ decisions, and requested Meta remove Amazon from the Muse experience.

Sources
  1. 01Ars Technica — Technology Lab — Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day
Also on Safety

Stories may contain errors. Dispatch is assembled with AI assistance and curated by human editors; despite the trust-score filter, mistakes happen. We correct publicly — every article links to its revision history. Nothing here is financial, legal, or medical advice. Verify before relying on any claim.

© 2026 Dispatch. No ads. No sponsorships. No paid placement. Reader-supported via Ko-fi.

Built by a person who cares about honest AI news.