 **Meta’s AI agent snoops on private messages and spits out other people’s bank balances - report**

A privacy nightmare is unfolding around Meta’s fast-growing [AI assistant,](https://t.me/geopolitics_prime/75018) which has already amassed 10 million downloads.

Muse is accessing private conversations and exposing personal data across accounts, according to internal company reports cited by Reuters.

**The AI agent:**

  accesses encrypted personal messages

  fabricates transaction details; serves up seemingly other users’ financial information

  sends unsolicited chats containing details people never shared with it

Meta says Muse only uses permissions users grant and that many of the odd messages are AI “hallucinations.” However, amid the [recent uproar](https://t.me/geopolitics_prime/74844) over “rogue” [AI systems,](https://t.me/geopolitics_prime/74714) employees are reportedly less reassured, with some describing the incidents as possible cross-account data leaks.

Muse is far from [the only culprit](https://t.me/geopolitics_prime/75902)amid a worrying pattern as tech companies push to give AI agents more control over everyday life.

** **[Substack](https://geopoliticsprime.substack.com/)** | **[Chat](https://t.me/+HLrnOk92f31mYzIy)** | **[@geopolitics_prime](https://t.me/geopolitics_prime)