Skip to main content
Mallory
← Back to News
ai-platform-securityendpoint-software-vulnerabilityidentity-authentication-vulnerabilitycredential-access-method

Meta’s Muse AI Agent Faces Message-Access and Account-Takeover Concerns

Emerging Technology Risk · 3 sources · Updated 5d ago · First seen Nov 10, 2026

Meta’s Muse AI Agent Faces Message-Access and Account-Takeover Concerns

Meta’s Muse AI agent faces security and privacy concerns following reports of unexpected access to private messages and a now-patched account-takeover vulnerability. Journalist Jason Aten reported that Muse accessed his Mac’s local Messages database despite his withholding explicit permission. Reporting described synchronization of 187,000 database rows while Full Disk Access was disabled. The concerns accompany Muse’s reported expansion to Apple Watch and Android WearOS, raising questions about connecting the agent to financial and other sensitive accounts despite assurances that users can approve actions before execution.

Researcher Patrick Wardle found that a locally modifiable speech-recognition server setting could redirect voice requests and authorization tokens to an attacker, enabling remote control of Muse within its existing permissions. Meta reportedly patched the flaw. Separately, the company reportedly ended an internal calling experiment that used human operators without adequately informing participating employees. Users should update Muse, minimize permissions, require confirmation for actions and website access, and disable personal-data use for model training. The agent’s App Store warning also advises users to monitor its activity and intervene because it may make mistakes or take unexpected actions.

New surface, new adversariesTrack how attackers are adapting to this technology.
Track this risk
Event timeline

How this story unfolded

Events from the most recent confirmed update back to the earliest known activity.

7 events
  1. Reported Nov 12, 20265d ago

    Meta ended the human-operator experiment after criticism

    Meta ended the experiment, acknowledged that conducting it without notifying participants was a mistake, and promised to revise the feature.

    Возможности и уязвимости в ИИ-агенте Muse* | Блог Касперского
  2. Reported Nov 12, 20265d ago

    Meta tested human operators handling Muse calls without adequate disclosure

    Meta conducted an internal employee experiment in which trained call-center operators handled Muse’s outgoing calls because recipients often hung up on automated voices. Many participating employees reportedly did not realize humans were handling their requests.

    Возможности и уязвимости в ИИ-агенте Muse* | Блог Касперского
  3. Reported Nov 12, 20265d ago

    Meta reportedly patched Muse’s token-theft vulnerability

    Meta reportedly released an update fixing the vulnerability discovered by Wardle. The source provided neither a CVE identifier nor a fixed-version number.

    Возможности и уязвимости в ИИ-агенте Muse* | Блог Касперского
  4. Reported Nov 12, 20265d ago

    Patrick Wardle discovered a Muse token-theft and remote-control vulnerability

    Wardle found that a local application or malicious Terminal command could redirect Muse’s speech-recognition traffic to an attacker-controlled server, exposing an authorization token. The token could enable persistent remote control of the agent within permissions already granted by the user, with more than 50 commands identified.

    Возможности и уязвимости в ИИ-агенте Muse* | Блог Касперского
  5. Reported Nov 12, 20265d ago

    Meta employee described permission requirements after Messages access concerns

    A Meta employee stated that reading messages required both separate Messages permission and Full Disk Access. The reported explanation did not establish how Muse accessed Aten’s messages without his explicit authorization.

    Возможности и уязвимости в ИИ-агенте Muse* | Блог Касперского
  6. Reported Nov 12, 20265d ago

    Muse accessed a journalist’s private Messages data without explicit permission

    After Jason Aten installed Muse on his iPhone and Mac while withholding explicit Messages access, the agent referenced private conversations and was found to have synchronized data from the Mac’s Messages database. Synchronization reached database row 187,462, which does not establish how many messages were processed; another report stated that Full Disk Access was disabled.

    Возможности и уязвимости в ИИ-агенте Muse* | Блог Касперского
  7. Reported Nov 10, 20268d ago

    Muse became available on Apple Watch and Android WearOS

    Muse was released for Apple Watch and Android WearOS, allowing users to interact with the agent through voice, text, and dictated messages.

    Muse di Meta arriva sugli smartwatch: perché non installarlo
Linked entities

Related entities

Vulnerabilities, threat actors, malware, products, organizations, and breaches Mallory has linked to this story.

Open in app →

Organizations

1 linked
Meta Platforms
Do you run this?Follow how adversaries are adapting to this technology, and where it touches your stack today.
Track This Risk
Mallory platform

See the full picture, correlated to your attack surface.

This page covers what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which threat actors are using it right now, and what to deploy next.

Exposure mappingMap indicators from this story to your assets and identify affected systems in minutes.
Threat actor evidenceEvery observed campaign, victim, and pivot linked to actors named in this story.
Associated malwareMalware, exploits, and IOCs connected to the activity described here.
Detection signaturesYARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Scheduled alertsGet matching new stories delivered to your team as they break, not the next morning.
AI threadsAsk questions about this story and take action on the answers.