Guide 17 · Safety

The Muse Mac Security Flaw: What Actually Happened

A researcher found a serious Mac app flaw; Meta patched it within a day. The full timeline and what it means for you.

By Abhit Tomar · Updated November 17, 2026 · 17 min read

Editorial illustration of a magnifying glass over a laptop revealing a cracked padlock
Illustration: museaicodes

Short answer

On September 21, 2026, researcher Patrick Wardle disclosed a zero-day in the Muse Mac app that could let local malware hijack the agent; Meta issued a hotfix on September 22. Keep the app updated.

The timeline

September 21, 2026: macOS security researcher Patrick Wardle published an X thread warning users about serious flaws in the Muse Mac app, alongside a proof-of-concept repository he titled 'not-a-mused.' His opening line: 'Please don't install — it's trivial to turn Muse into the ultimate backdoor.'

September 22, 2026: Wardle posted 'Hooray, hot-fixed!' confirming Meta had patched the vulnerability — roughly a day after disclosure. David Singleton of Meta Superintelligence Labs confirmed the hotfix on X, describing the issue as a local privilege escalation attack, not a remote exploit, and assessing the practical risk to users as 'quite low.'

What the flaw was

The vulnerability centered on an undocumented preference setting in the Mac app — endo_voyager_dictation_endpoint — which controlled where Muse sent voice dictation for processing. Any app or script running under the user's account could modify this setting without triggering macOS permission alerts, redirecting transcription from Meta's servers to an attacker's endpoint.

That redirection exposed the user's account authentication token. With the token, an attacker gained full control of the agent — and because Muse holds broad system permissions by design, Wardle's proof of concept could take pictures and write malicious files to disk, often without alerting the user. As he told Ars Technica: 'We can manipulate the agent and leverage its privileges to do whatever we want. So instead of us having to write a very comprehensive Mac malware stealer, we can just leverage the AI assistant itself.'

Several design choices enabled it: dictation processed in the cloud rather than on-device (unlike Apple's on-device transcription), and undocumented settings left writable by any local process.

Why could any local process rewrite that setting at all? On macOS, an app's preferences live in files under the user's own account, and anything running as that user can normally read and write them. macOS permission prompts protect specific things — the camera, the microphone, your folders — but an app's own preference files are not on that protected list, because apps are expected to manage their own settings. Wardle's finding was that Muse had put a security-critical choice — where voice audio gets sent — in a place any local process could reach, instead of treating it as a protected setting. The flaw was not that macOS failed; it was that the app trusted a writable file with a decision that should have been locked down.

How serious was it, really

Two things are true at once. The exploit required local code execution — malware already running on the machine, physical access, or a social-engineering trick like a ClickFix prompt getting the user to paste a terminal command. It was not remotely exploitable over the internet, which is why Meta assessed real-world risk as low.

But the deeper point stands: an agent designed to act across your files, email, messages, and calendar with broad permissions becomes the single highest-value target on the machine. Compromise the agent and you inherit everything it can touch. That structural reality doesn't disappear with one hotfix — it's the permanent trade-off of computer-use agents. Our privacy guide answers what Muse can see and what you can opt out of.

Meta's 'quite low' risk assessment and Wardle's 'ultimate backdoor' warning can both be true, because they measure different things. Likelihood was low: the attack needed malware already on the machine, and most Mac users never get that far. Impact was high: once the agent was hijacked, the attacker inherited everything Muse could touch — files, messages, email, calendar — without writing a dedicated stealer. That gap between likelihood and impact is the right way to think about agent security generally. A flaw that needs local access sounds minor until you remember that the whole point of an agent is to concentrate your permissions in one place, so any single hole in it pays off more than the same hole in an ordinary app.

What to do as a user

  • Update the Muse Mac app and keep it updated — this is how security fixes reach you.
  • Never paste terminal commands from strangers, popups, or videos. ClickFix-style tricks are the cheapest attack path.
  • Keep macOS itself updated; OS-level protections are part of the defense.
  • Grant Muse the narrowest permissions each task needs, and revoke access you no longer use.
  • Treat any agent with deep system access as high-trust software: powerful, useful, and worth a skeptical eye.

The bigger picture

This episode landed in a week when platforms started pushing back on autonomous agents more broadly — Amazon began blocking Muse from placing automated orders, saying the agent violated its terms. The industry is negotiating, in real time, what agents are allowed to do and who is responsible when they do it.

Our honest take: Muse's capabilities are real, and so are the risks that come with an agent holding your credentials and permissions. Use it, but grant least privilege, keep everything updated, and remember that the most powerful assistant on your machine is also the most attractive target on it.

The pattern here is bigger than one setting in one app. Every agent that acts across your files, inbox, and calendar is a new kind of target: compromise the agent and you skip the work of compromising each app separately. That is why security researchers are now probing agent permissions the way they used to probe browsers — and why the fixes that matter are structural, not one-off. Expect more of these disclosures as agents spread, and judge each one the same way: what did the attacker need first, what did the agent hand over, and how fast was the fix.

What responsible disclosure means in practice

When a security flaw in an AI product is reported responsibly, there is a standard sequence behind the scenes. A researcher finds the vulnerability and reports it privately to the company instead of publishing it, giving the company time to investigate and fix the problem before attackers learn about it. The company typically confirms the report, develops a patch, rolls it out, and only then does the researcher publish the details — often after a coordinated date both sides agreed on. This is why you sometimes hear about a flaw weeks or months after it was first found; the delay is the system working as intended. Public disclosure after a fix serves everyone: users learn what happened and what to do, researchers get credit, and the pressure to fix real issues stays on. Be skeptical of flaw reports that arrive with no fix, no timeline, and no named researcher — those deserve verification before action.

What to do right after a flaw is reported

Your response should be proportionate and quick. First, update the app to the latest version through your official app store — if a patch exists, that is the single most important step. Second, if the flaw involved account access, sessions, or authentication, change your password and sign out of all sessions, then sign back in. Third, review the third-party apps and services connected to your account and disconnect anything you do not recognize or no longer use; a flaw's blast radius is limited by how many doors into your account exist. Fourth, check whether the company published any specific guidance for the flaw in question — sometimes there are extra steps, like regenerating API keys or re-verifying a device. Do these in order, calmly, the same day you learn about the issue. Panic-deleting accounts or data usually causes more harm than the flaw itself.

How to check whether you were affected

Companies typically say whether a flaw was exploited in the wild or only demonstrated in a lab, and that distinction matters enormously. A lab-only demonstration means the risk to you was theoretical; active exploitation means you should assume exposure and act accordingly. Look for the company's official statement — on its security page, blog, or status channel — rather than relying on social media summaries, which tend to inflate. Check your account's recent activity or login history if the product offers it, looking for sessions or actions you do not recognize. If the company offers a notification or lookup tool for affected users, use it. And keep an eye on your email for official communications, while being careful about phishing: attackers love to send fake "security alert" emails right after a real flaw makes news, so verify any links by navigating to the company's site yourself.

Security hygiene going forward

A single flaw is a reminder to keep the basics in shape, not a reason to live in fear. Use a unique password for your AI accounts, ideally stored in a password manager, and turn on two-factor authentication wherever it is offered. Keep apps updated automatically so patches reach you without effort. Periodically audit connected apps, active sessions, and devices. Be cautious about what you paste into AI tools — sensitive documents, credentials, and private data should be shared sparingly regardless of any flaw, since you cannot fully control what happens to inputs once sent. Finally, follow the product's official security communications channel so you hear about future issues from the source rather than from rumors. Good hygiene does not make you invulnerable, but it shrinks every future incident down to a manageable size.

How to check whether your Mac app is patched

The fix shipped on September 22, 2026, so any install or update from that date onward includes it. The simplest check is to open the App Store, look for Muse under updates, and install anything pending — then confirm automatic updates are on so the next fix reaches you the same way. If your copy came from anywhere other than the official listing, replace it with the official one rather than trusting a copy that has been sitting on your machine since launch week. There is no version number to memorize and no setting to toggle: patched is the default state of a current install.

Is Muse safe to use now?

Yes, with the same caveats that apply to any powerful agent. The specific flaw is patched, the attack required local malware rather than a remote exploit, and Meta fixed it within a day of disclosure — all of which are good signs. What has not changed is the structural trade-off: Muse holds broad permissions by design, so it remains a high-value target, and future flaws are a matter of when, not if. The sensible posture is to keep using it if it is useful, keep it updated, grant the narrowest permissions each task needs, and never paste terminal commands from strangers. That combination is what 'safe enough' looks like for an agent.