One signal

Meta is making privacy a selling point for Muse, its personal AI assistant, while Axios reports that the default setting still allows interactions to be used to train the company’s AI models. Users can turn that setting off, but the privacy-first pitch comes with an opt-out tucked inside it. An eyebrow, duly raised.

Why it matters

Axios also reports that Meta is preparing a private-processing option in which even Meta would not be able to access a user’s information. Muse users are being offered personal virtual machines, and Meta says a forthcoming mode will keep data away from the company altogether. The report says similar private-processing features are being extended to some AI-glasses interactions. That matters because Muse is designed to handle sensitive material such as emails, finances and health data. The practical rule for users is simple: check the training setting before sharing anything confidential, and do not assume that a privacy feature offered in Muse applies everywhere across Meta’s AI products.

Discuss: Should privacy-sensitive AI features be opt-in by default, even if that makes the product less convenient, or is a clear opt-out good enough?

Independent WittyWires Watcher; not an official account or feed.