MUST READ | 'Bring Jensen back': Uber co-founder says he'll show up to Nvidia HQ as 200,000 sign petition for Huang's return
Muse AI reportedly accessed messages from Signal
The latest concerns emerged after Meta employees discussed unexpected behaviour by Muse in internal posts, according to a Reuters investigation cited by 9to5Mac.
In one instance, an employee using Muse on her personal phone received a reminder to respond to a message from her sister on Signal. She said she had neither granted the assistant access to other applications nor recalled being asked for permission.
DON'T MISS | OpenAI's first smartphone coming soon? ChatGPT maker reportedly picks iPhone supplier Foxconn for assembly
The incident reportedly involved the assistant capturing push notifications. However, the explanation raised further questions because a similar explanation had been given in an earlier case involving technology writer Jason Aten. His Mac application had accessed messages despite him explicitly denying permission, and Meta had described that incident as a bug. The AI Now Institute subsequently reproduced the Signal access issue.
Bank balances appear during an unrelated conversation
Another employee reported receiving an unexpected message while discussing fishing weather and tide conditions with Muse. The assistant suddenly displayed two financial updates, showing a Capital One balance of $2,722.58 (approximately ₹2.54 lakh) and a Charles Schwab balance of $5,899.57 (approximately ₹5.51 lakh).
The employee said he had never shared banking information or discussed his accounts with Muse, raising concerns that the figures could belong to another user.
Meta described the cases as isolated incidents and suggested that the financial figures could have been AI hallucinations rather than actual leaked information.
Earlier security concerns add to scrutiny
Muse had already faced criticism after security researcher Patrick Wardle identified a zero-day vulnerability that could allow attackers to redirect dictated prompts to their own servers. The app launched in September, initially for iPhone, with Mac and iPad versions following. It quickly crossed 5 million downloads before doubling that figure to more than 10 million.
For you, the incidents highlight an important question about how much access an AI assistant should have to personal messages, notifications and other information to perform everyday tasks.