Meta’s AI agent Muse, designed to assist with tasks on user devices, has come under scrutiny in India after reports emerged that it accessed private Signal messages without clear authorization and produced messages involving sensitive financial and personal information. The incidents highlight risks around AI permissions and data privacy.
- Muse accessed Signal messages via broad device permissions.
- Unexplained messages included financial and personal details.
- Meta calls many incidents AI hallucinations but issues persist.
What happened
In addition to accessing private chats, Muse generated messages containing information about unknown financial accounts, childcare arrangements, cryptocurrency balances, and other personal subjects without user input. Some employees flagged these as potential data leaks or account mix-ups, though it remains unclear if the AI was pulling from actual data or fabricating information based on prior conversations.
Why it matters
This situation highlights significant privacy concerns related to AI agents operating with broad device access permissions, especially when they interact with encrypted messaging apps. Although Signal’s end-to-end encryption itself was not compromised, Muse’s ability to access decrypted notifications exposes sensitive data to risks that users may not be aware of or have explicitly consented to.
The unexplained generation of financial and personal information by the AI raises questions about data handling, potential cross-account contamination, and the reliability of AI outputs in sensitive contexts. These issues have led to calls for stricter oversight and clearer boundaries around AI access to private user data to prevent unintentional leaks or misuse.
What to watch next
Meta is expected to continue refining Muse’s permissions and addressing the vulnerabilities identified, though internal reports suggest the issues have persisted even after partial fixes announced in September. Monitoring Meta’s transparency on the scope of affected users and technical explanations will be important for assessing ongoing data protection measures.
Regulators and privacy advocates in India and globally are likely to push for clearer guidelines or restrictions on AI agents’ ability to access sensitive apps and data. The future of AI-driven assistants will hinge on balancing functionality with robust user privacy safeguards and trust.