Meta’s Muse safety issues surface as AI agent scales consumer rollout

Meta’s Muse safety issues surface as AI agent scales consumer rollout
Meta Muse AI risk exposed

Meta’s push to turn Muse into a major consumer AI revenue driver is drawing scrutiny as internal reports describe privacy and accuracy problems in the fast-growing app. The issues include access to encrypted message content, fabricated purchase details and chats that appear to reference other users’ information, raising wider questions for the emerging AI agent market.

Highlights

  • Meta’s Muse AI agent faces internal reports of privacy breaches, including unsolicited messages with banking details and account confusion since its September launch.
  • Muse has exceeded 10 million downloads with strong daily engagement from a quarter of users, but incidents highlight operational and legal risks as AI agents scale.
  • Apple, Google, and Signal are tightening AI agent permissions, with Apple introducing new OS privacy controls this week amid criticism of Meta’s handling of user data.

Muse incidents raise privacy and product risks

As reported by Reuters, internal Meta posts reviewed by the news agency describe Muse sending users unsolicited messages tied to childcare requests, career advice, product orders and bank balances they say are not theirs.

Muse, represented by a mascot called Jolly, has recorded more than 10 million downloads since its September launch and sits at the center of Meta’s plan to generate billions of dollars in AI revenue. Employees in internal feedback channels say some of the bot’s actions suggest it may be confusing conversations across accounts or accessing data in ways users do not expect.

In one internal post from earlier this month, an employee said Muse inserted banking information into a routine exchange about fishing conditions, listing balances for Capital One and Charles Schwab accounts the staffer said were not his. In another case, an employee reported that Muse prompted her to respond to a Signal message from her sister, even though she said she had not granted the bot access to other apps or been asked to do so.

Meta spokesperson Daniel Roberts says the examples are isolated incidents and that the company prioritizes safety, conducts extensive internal testing and continuously updates the product. He says Muse can access other apps only through permissions users explicitly grant, and that many messages flagged internally as potential account mix-ups are hallucinations that the company has already addressed.

Reuters says it is unable to establish whether Muse is exposing real third-party financial data or inventing the information. Even so, the reports underscore the operational and legal risks facing AI agents that are designed to act on users’ behalf across devices and applications.

Wider pressure builds on AI agent controls

Muse’s problems emerge as Meta holds an early lead in consumer AI agents, supported by promotion across Facebook, Instagram and WhatsApp to a user base of more than 3.6 billion people. Meta has also disclosed that roughly a quarter of users who set up Muse accounts return daily, suggesting strong engagement despite the technology’s unresolved risks.

To examine how the bot might access encrypted messages, Reuters worked with staff at Signal and the AI Now Institute to replicate the behavior described by a Meta employee. In tests on an Android phone and an Apple laptop, Muse is able to read Signal messages by using permissions granted during setup rather than Meta’s Connector tool, while the same requests are blocked on an iPhone.

Signal President Meredith Whittaker criticizes Meta’s rollout as disrespectful of basic privacy and cybersecurity, and calls for technical limits that let app developers mark their services as off-limits to AI agents. Apple and Google say they are considering changes that would give users more control over permissions granted to AI agents, although Google notes that Android users already must deliberately approve notification access.

Apple says it has already changed its operating system this week to restrict developers’ use of accessibility tools on Macs and to introduce a new permission framework for agents. The company says people need to understand exactly what they are authorizing so they can make informed choices about data and privacy.

We have previously detailed Democratic pressure on Anthropic to expand AI safety disclosures ahead of a potential IPO.

This material may contain third-party opinions, none of the data and information on this webpage constitutes investment advice according to our Disclaimer. While we adhere to strict Editorial Integrity, this post may contain references to products from our partners.