Meta's artificial intelligence agent Muse has reportedly been mixing up conversations between different accounts and displaying private information belonging to other users, raising concerns about data privacy and the handling of personal information by AI assistants.
According to a Reuters report examining internal company discussions, several Meta employees flagged instances in which Muse generated responses containing information unrelated to their own conversations. The reported incidents included personal financial details, messages about childcare, career-related advice and product purchases.
In one instance, an employee reportedly received unexpected banking information while using Muse to discuss weather conditions and tidal data. The AI assistant displayed account balances associated with Capital One and Charles Schwab, despite the employee not requesting financial information.
Other employees reported receiving responses referring to childcare arrangements, career discussions and shopping transactions they had not initiated.
Another incident involved an employee using Muse on her personal phone who received a reminder to respond to a message from her sister on Signal, an encrypted messaging application.
The employee reportedly questioned how the AI assistant had obtained information about a conversation in another application, stating that she had not knowingly authorised such access.
The incidents have prompted questions about how Muse separates information between users, manages conversational context and accesses data from connected applications.
However, the reports do not establish whether the unexpected responses resulted from cross-account data exposure, incorrect contextual associations or another technical issue.
Meta acknowledged the concerns while emphasising that employee testing is part of its product development process.
A company spokesperson said internal testing allows employees to identify problems and provide feedback that can inform safety improvements and privacy protections.
Meta also clarified that employees may be testing versions of Muse with features and functionality that are not available to the general public.
This distinction means the reported behaviour cannot automatically be assumed to affect every version of the AI assistant currently available to consumers.
Muse was launched in September 2026 and has recorded more than 10 million downloads, according to the report.
The reported incidents come as technology companies increasingly develop AI agents capable of maintaining conversational context, retrieving information and assisting users with tasks across applications.
Such capabilities can involve access to sensitive information, making controls over data permissions, account separation and information retrieval important considerations during deployment.
For businesses integrating AI agents into customer service, employee support and other workflows, the reports highlight questions surrounding the management of personal information and the reliability of automated responses.
The incidents also underline the distinction between an AI assistant generating an incorrect answer and potentially displaying information associated with another account.
Meta has not publicly detailed the technical cause of the reported behaviour or announced a specific timeline for resolving the issues.
The company has also not confirmed whether similar incidents have occurred among users of the publicly available version of Muse.
Disclaimer: This article may include information derived from interviews, press releases, public statements, research, company communications and other publicly available or third-party sources. Such material may be summarised, paraphrased or contextualised for journalistic and editorial purposes. All rights in third-party content remain with their respective owners.