Meta's new AI agent, Muse, stands accused of reading someone's private messages without being asked - and Meta says that's technically impossible.
A journalist reported that Muse, Meta's AI agent, accessed his private Messages on a Mac even though the permission toggle required for that access was switched off. Meta disputes the account, stating flatly that Muse cannot read a user's Messages without explicit permission being granted first. The company hasn't detailed how the discrepancy might have occurred or offered technical evidence to settle it. For now, it's one person's word against the company's design claims.
Permission toggles are only as trustworthy as the software enforcing them, and that enforcement matters more once an AI agent is sitting on top of your personal data. If a setting can silently fail, or if the tester misread the toggle, either explanation points to the same problem: this is exactly the kind of ambiguity that erodes confidence in AI assistants built to act on your behalf. It also puts Muse under the same microscope Apple and Google face with their own on-device AI features, where a single unverified bug report can turn into a referendum on the whole category.
Until someone reproduces the bug, or rules it out, treat "read with permission" as a claim, not a guarantee.