Osegi
Mobile·2 min read

Muse AI privacy problems: Meta staff report bank balances

Meta's own employees say the Muse AI app read a Signal message it had no permission for and sent one of them what looks like another user's bank balances.

An iPhone showing the Meta Muse AI app with a padlock icon, suggesting privacy concerns.
Image: 9to5Mac

Meta's Muse AI app has a privacy problem, and this time the people raising it work at Meta. Staff have described two incidents on internal posts, including one where Muse apparently handed an employee someone else's bank balances.

Muse is currently the second most popular free iPhone app in the App Store, with more than 10 million downloads. Meta launched it in September as "your personal AI agent that gets things done across every part of life." It started on iPhone, with Mac and iPad apps soon after, and it quickly overtook ChatGPT on the US charts after a big advertising push.

What did the Muse AI app do?

According to Reuters, one employee using Muse on her personal phone got a reminder to reply to her sister on Signal, the encrypted messenger that competes with Meta's WhatsApp. "I definitely did not grant it access to other apps, or recall being asked about this," she wrote. The report says this appears to have happened through captured push notifications. The AI Now Institute was able to reproduce the problem.

A second employee was chatting with Muse about fishing weather and tides when it added: "BTW, two bank updates, your Capital One balance is $2722.58, your Charles Schwab balance is $5899.57." He wrote that he had never shared a bank account with Muse or discussed banking, and feared it was sending him another person's details.

Not the first warning

This follows earlier trouble. Tech writer Jason Aten said the Mac app read his messages after he explicitly refused permission, then gave him a false explanation of how. Meta called that a bug. Security researcher Patrick Wardle also found a zero-day, a flaw unknown to the vendor, that could let an attacker redirect dictated prompts to their own server.

The notification explanation was also given in Aten's case, and it turned out not to be true.

Meta told Reuters these were "isolated incidents" and suggested the bank figures may have been AI hallucinations, meaning invented output, rather than leaked data. That would still mean the app told a user made-up financial details as if they were real.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Tags:metamuse aiprivacysignaliphoneapp store

More from this day

See all →

More in: Mobile

See all →