Meta has introduced privacy controls for its AI assistant Muse that let users choose which apps the assistant connects to and opt out of having their interactions used to train Meta’s models. The company also said Muse data will not feed Meta’s advertising systems, and it is working on a more ambitious confidential VM option that would be encrypted with a key held only by the user.
Why this matters
For users deciding between many capable models, clarity about how data are treated is a decisive factor—especially for sensitive information. The article’s author says that Meta’s clearer privacy options have made them willing to tentatively resume using Muse and Meta’s AI-enabled glasses.
What Meta has said so far
In a 6,500-word August manifesto, CEO Mark Zuckerberg articulated the company’s commitment: “People should have a fully private mode for personal agents where even Meta or any other service provider cannot see or grant access to your information.” At Muse’s launch Meta provided settings to control app connections and to prevent interactions from training its systems. The company also said it plans to bring private processing to AI glasses by year-end, protecting data even while processed in the cloud, though it did not give a specific date.
Early user experience
The author began cautiously trying Muse: they turned off the setting that allows Meta to use their interactions to train systems and are awaiting the confidential option. For now they use Muse for tasks they do not mind Meta knowing about—or that Meta likely already knows.
Examples given include asking Muse to notify them when the Golden State Valkyries’ first playoff game time was announced; the author woke on Friday to a Muse notification that the schedule had been released. They also asked Muse for recommendations for a removable, portable CarPlay display for a street-parked car and received several reasonable suggestions.
The author has been an early user of Meta’s glasses since the first version (which initially had little in the way of AI). They valued the glasses’ cameras for recording sports, pets and children—moments better preserved by staying present than by pulling out a phone. Over time the glasses gained AI features, but the prospect that uploaded photos or other data could be used for ad targeting and other purposes was a barrier to continued use.
Remaining reservations
The author frames their approval as tentative: it is based on what Meta says it will do, and the company’s privacy record is mixed. The piece references a New Mexico jury verdict on Friday without detailing it. Zuckerberg has said Meta will take a cut of transactions made via Muse, allowing the company to offer the service free for most customers.
The author also recalls Zuckerberg’s earlier pledge to deploy end-to-end encryption across Meta’s messaging platforms and subsequent retreat from some commitments. Additional privacy concerns remain unaddressed by Meta’s recent announcements, particularly regarding the rights of bystanders whose likenesses may be captured by Ray-Bans I and other devices embedding microphones and cameras.
Alan Butler, executive director of the Electronic Privacy Information Center, is quoted in the article: “Meta's products that embed microphones and cameras in everyday devices pose serious privacy risks.”
Bottom line
The author is willing to experiment with Muse and Meta AI on their glasses but maintains caution: they will trust, but verify, and are not ready to cede full control over their data.
