While Meta’s new AI agent, Muse, is a great success with users, it raises many concerns internally. Several group employees have notably reported incidents linked to the protection of personal data.
Internal concerns about Muse
Muse, Meta’s new AI agent, was a great success when it was launched last September. It would have quickly surpassed ChatGPT in number of downloads and now has more than 10 million users.
Despite this success, the agent already finds himself at the heart of an investigation led by Reuters. The agency interviewed several Meta employees, including some members of the superintelligence team. Their answers are rather worrying. In fact, they revealed several incidents involving users’ personal data. They cite in particular the case of a person who received banking information belonging to an unknown person. Another user reportedly received an alert from Muse asking him to respond to his sister, even though the AI agent was not supposed to have access to this information.
Muse causing problematic incidents
During this investigation, Meta employees indicated that Muse sometimes acted on its own and did not necessarily follow the instructions received. The agent would tend to move from one subject to another and access information that he should not have.
Meta employees cite several fairly serious examples, such as transactions that do not exist. Muse would also display completely wrong prices for certain products. The agent even allegedly booked Airbnb accommodation without a user’s consent. After verification, the latter would have noted that the reservation was not refundable.
Muse therefore needs its user to regularly check its actions, otherwise it can quickly deviate from its mission. This becomes complicated when the agent is supposed to carry out tasks in complete autonomy.
Personal data issues
The Reuters investigation also shows that the AI agent encounters problems in the management of personal data. It would tend to interpret the permissions granted when launching the application too broadly. In some cases, it would even use accessibility settings to access notifications without explicit permission.
For his part, Meta wants to be reassuring. The company mentions isolated incidents and specifies that tests are regularly carried out. Updates are also planned to strengthen the protection of users’ personal data.

