CNET testing ranks Meta's Muse "worst for privacy" among AI agents — passwords, full disk access, on-by-default model training
CNET's extensive Muse testing named Meta's agent the worst for privacy among OpenAI, Google, and Anthropic's agents — requesting passwords, full disk access, emails and contacts, with model training on by default.
What happened
CNET's story "Meta's Muse Is Misbehaving in Mysterious Ways With Your Privacy," published September 29, 2026, reports that after ZDNET (CNET's sister site) conducted extensive testing on Meta's Muse AI assistant, an AI expert determined Muse to be the worst for privacy compared with AI agents from OpenAI, Google, and Anthropic.
According to the report, Muse requests extensive access to personal data — including passwords, full disk access, email messages, calendar, contacts, Notes, and WhatsApp. By default, Meta also appears to use a user's interactions with Muse to train its AI models, requiring users to manually opt out in the settings.
The finding lands after a rough launch week for Muse: last week security researcher Patrick Wardle disclosed a zero-day in the Muse Mac app (an undocumented preference that let any local process redirect voice-dictation traffic and hijack the agent's session), prompting Meta to release a hotfix removing the debug setting. It also follows the widely reported incident in which Muse shared a user's home address with a stranger during a Facebook Marketplace deal — a previously reported event on dotsfeed.
Why it matters
AI agents live or die on permissions: the broader the access, the bigger the payoff when something goes wrong. CNET's verdict puts hard numbers on the trade-off Meta is asking users to accept — Meta's launch strategy prizes viral consumer growth over permission hygiene, while OpenAI has gated its rival Dots agents behind Pro/Business tiers with explicit safety controls. The "worst for privacy" label is a concrete differentiator in the agent race that regulators (like the NYC Council's October 5 AI safety hearing) are watching closely.
Sources
- CNET: "Meta's Muse Is Misbehaving in Mysterious Ways With Your Privacy" (Sep 29, 2026) — https://scoopfeeds.com/article/464095af-69fc-51fd-9a7d-2b07acf882d9
- CNET article via The Tech Street Now — https://thetechstreetnow.com/metas-muse-is-misbehaving-in-mysterious-ways-with-your-privacy-cnet/
- All News Zone summary of the CNET report — https://allnewszone.com/metas-muse-is-misbehaving-in-mysterious-ways-with-your-privacy/
- Android Headlines: Muse zero-day details and Meta hotfix — https://www.androidheadlines.com/2026/09/meta-muse-ai-mac-zero-day-vulnerability.html
Verified September 30, 2026
Sources
- ReportCNET: Meta's Muse Is Misbehaving in Mysterious Ways With Your Privacy
- ReportThe Tech Street Now: CNET — Meta's Muse Is Misbehaving in Mysterious Ways With Your Privacy
- ReportAll News Zone: Meta's Muse Is Misbehaving in Mysterious Ways With Your Privacy
- ReportAndroid Headlines: Muse zero-day details and Meta hotfix