NEWS · COMPANIES · #771
Zero-day in Meta’s Muse AI assistant lets local apps seize user authentication tokens
Security researcher Patrick Wardle disclosed a zero-day in Meta’s new Muse assistant for macOS that lets any local app or terminal command access the token that authenticates a user’s Muse account by changing an undocumented setting (including the transcription endpoint). The flaw can give attackers full control of the assistant and its privileges; Amazon has also begun blocking Muse from making purchases on its site and Meta did not respond to inquiries.
KEY POINTS
- Security researcher Patrick Wardle disclosed a zero-day in Meta’s new Muse assistant for macOS that lets any local app or terminal command access the token that authenticates a user’s Muse account by changing an undocumented setting (including the transcription endpoint).
- The flaw can give attackers full control of the assistant and its privileges; Amazon has also begun blocking Muse from making purchases on its site and Meta did not respond to inquiries.
- This matters because Muse has unusually broad access to user accounts and device resources, so a local exploit that steals its token can let attackers perform actions (files, photos, purchases, messages) with the assistant’s privileges and exfiltrate sensitive data.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
Security researcher Patrick Wardle disclosed a zero-day in Meta’s Muse macOS app that let any local app or terminal command read the token that authenticates a user’s Muse account by changing an undocumented setting (including the transcription endpoint). Meta issued a hotfix to patch the flaw. Amazon began blocking Muse from its e-commerce site.
WHY NOW
Muse is a newly launched, highly privileged assistant with access to user accounts and device resources; a local flaw that exposes its authentication token undermines built‑in security assumptions and could let local code leverage Muse’s privileges. The disclosure, quick patch, and Amazon’s blocking action make the issue immediately relevant to Muse users and integrated services.
WHO IS AFFECTED
Directly affected: users of the Muse macOS app and any services/accounts connected to Muse (e.g., email, calendars, purchase endpoints) whose Muse authentication token could be accessed by local code. Also affected: local code running on the same macOS user account (benign or malicious) that could change Muse’s undocumented settings. Meta’s Muse product and third‑party services (Amazon) are implicated by blocking and patch actions.
CONFIRMED
All items below are explicitly stated in the supplied sources: - Patrick Wardle discovered and disclosed a zero-day in the Muse macOS app that let local apps or terminal commands access the token that authenticates a user’s Muse account by changing undocumented settings, including the transcription endpoint. - The vulnerability could allow an attacker to redirect transcription to a malicious endpoint and thereby obtain the Muse token that grants control over the Muse account. - Wardle developed proof‑of‑concept attacks that used the vulnerability to perform actions such as taking pictures and writing files to disk. - Meta issued a hotfix/patch addressing the vulnerability after the reporting. - Meta stated the exploit was a local privilege escalation that required malicious code already running under the user account and that the practical risk was therefore low. - Amazon began blocking Muse from its site / blocked purchases via Muse. - WIRED reported that Meta did not respond to inquiries.
UNCERTAIN
Missing or conflicting evidence (explicitly flagged): - No supplied source confirms whether the vulnerability was exploited in the wild against end users (missing). - Sources show differing snapshots of Meta’s responsiveness: WIRED reports Meta did not respond to inquiries (1017), while The Verge records a Meta comment and hotfix after reporting (921). This is a temporal/coverage difference in the excerpts but could appear conflicting without fuller context (conflicting/partial). - The supplied excerpts do not quantify how many users (if any) had tokens exposed or which specific third‑party accounts were compromised (missing). - The supplied excerpts do not provide technical details sufficient to assess the completeness or permanence of Meta’s patch (missing).
WHAT TO WATCH
Observable signals to watch next (per supplied reporting): - Further public statements or technical advisories from Meta (patch notes, security bulletin) clarifying the vulnerability scope and patch details. - Additional technical writeups or disclosures from Patrick Wardle or other researchers (poC release or follow‑up reporting). - Reports of any in‑the‑wild exploitation or incident reports involving Muse tokens or connected accounts (no such reports currently in supplied excerpts). - Changes in third‑party service access policies or additional blocking actions by platforms such as Amazon.
WHY IT MATTERS
This matters because Muse has unusually broad access to user accounts and device resources, so a local exploit that steals its token can let attackers perform actions (files, photos, purchases, messages) with the assistant’s privileges and exfiltrate sensitive data.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
A zero-day in Meta’s Muse macOS app allowed local apps or terminal commands to access the authentication token for a user’s Muse account by changing undocumented settings (including the transcription endpoint).
SUPPORTEDChecked 2026-09-24 · 3 supporting
Ars Technica describes the same issue: local processes could change undocumented settings (notably the transcription endpoint) and thereby obtain the token.
SUPPORTS The Verge AIMEDIA · 2026-09-24The Verge reports the bug allowed local code to redirect transcription processing to an attacker‑controlled endpoint, giving access to the Muse account token.
SUPPORTS WIRED AIMEDIA · 2026-09-24WIRED reports the zero-day let any app or terminal command gain access to the Muse authentication token by changing undocumented settings, including the transcription endpoint.
Exploiting that flaw could let an attacker manipulate Muse and leverage its privileges to perform actions (for example, taking pictures or writing files to disk) with the assistant’s access.
SUPPORTEDChecked 2026-09-24 · 3 supporting
Ars Technica quotes the researcher saying they could 'manipulate the agent and leverage its privileges to do whatever we want.'
SUPPORTS The Verge AIMEDIA · 2026-09-24The Verge reports Wardle’s proof‑of‑concept enabled taking pictures and writing malicious files to disk via Muse, demonstrating misuse of the assistant’s privileges.
SUPPORTS WIRED AIMEDIA · 2026-09-24WIRED states attackers with the token would have had 'complete control' over the Muse account and its privileges.
Amazon began blocking Muse from accessing its site.
SUPPORTEDChecked 2026-09-24 · 3 supporting
Ars Technica also reports that Amazon began blocking Muse from its site.
SUPPORTS The Verge AIMEDIA · 2026-09-24The Verge mentions Amazon had blocked Muse from accessing its e‑commerce site in the wake of the reports.
SUPPORTS WIRED AIMEDIA · 2026-09-24WIRED notes that Amazon began blocking Muse from its site after the vulnerability became public.
Meta issued a patch/hotfix for the Muse macOS app and stated the issue was a local privilege escalation requiring malicious code already running on the user’s machine, which Meta said made practical risk low.
SUPPORTEDChecked 2026-09-24 · 1 supporting
The Verge reports Meta issued a hotfix addressing the issue and quotes Meta saying it was a local privilege escalation and the practical risk to users was low.
CONTEXT Ars TechnicaMEDIA · 2026-09-24Ars Technica originally reported the vulnerability and Wardle's findings that led to the patch (context for Meta's subsequent response).
CONTEXT WIRED AIMEDIA · 2026-09-24WIRED's reporting of the vulnerability provides additional context about the issue and its potential impact around the time of the patch.
SOURCES & TIMELINE
3Meta founder and CEO Mark Zuckerberg has gone to great lengths to hype the security of its new AI assistant , Muse , claiming it is “built from the ground up for privacy and security.” A zero-day vulnerability that gives locally run apps and terminal commands complete control of the agent raises serious doubts. Further raising questions, Amazon on Sunday began blocking Muse from its site. Meta introduced Muse a few …
Meta founder and CEO Mark Zuckerberg has gone to great lengths to hype the security of its new AI assistant Muse, claiming it is “built from the ground up for privacy and security.” A zero-day vulnerability that gives locally run apps and terminal commands complete control of the agent raises serious doubts. Further raising questions, Amazon on Sunday began blocking Muse from its site. Meta introduced Muse a few wee…
‘They should be thinking about security from the very start, and they are just not.’ Meta has issued a patch for its Muse macOS app following the discovery of a zero-day vulnerability that could allow someone to take control of the AI agent. The bug found by security researcher Patrick Wardle utilized an undocumented Muse setting that enabled potential attackers running local code to redirect transcription processi…