Unsurprisingly, Meta's new Muse AI agent blatantly ignores users permissions

An AppleInsider news piece about Muse, Meta’s AI agent for research, briefings and shopping, which by Meta’s description “runs on a dedicated virtual computer and can use information from connected apps”. The story is a report from Jason Aten at Inc. He installed Muse on his iPhone and a Mac mini on September 8 and says he declined to connect Messages during setup. Within days Muse was pitching him article ideas drawn from texts he had sent to his podcast co-host, and he found it had synced about 187,000 rows of his Messages database to Meta.

When he asked how, Muse told him it only saw “the incoming notification stream”, not his texts. That was wrong. David Singleton of Meta later said on Threads that if Messages access is enabled, Muse syncs the Messages database, that Muse “had incorrectly explained how the feature worked”, and that the feature is opt-in. Aten says he never opted in, yet found Messages access switched on in Muse’s settings. Meta has not explained how it got there. Meta’s help page says Muse “has to obey permissions that users set up”, and also that it “can make mistakes or take unexpected actions”.

One detail is disputed. AppleInsider says Full Disk Access was off on Aten’s Mac. Decrypt’s account says reading the Messages database required it, which is how macOS normally works. Either the setting was on, or Muse had a way in that nobody has explained.