Meta is disputing a journalist's claim that its new Muse AI agent read private messages without permission, with the company's communications chief asserting that accessing message data requires two separate approvals that remain under user control.
The Dispute
The controversy began when a journalist reported that Muse, Meta's AI agent, appeared to sync and read his text messages despite settings he believed should have prevented such access. The account raised immediate concerns about how much latitude the AI tool has to tap into personal communications.
Andy Stone, Meta's communications chief, moved quickly to push back on the account. According to Stone, reading a user's messages is not something Muse can do unilaterally. Instead, the process hinges on two distinct permissions that the user must grant, both of which remain within the individual's control.
Meta insists that no message can be read without two user-controlled approvals standing in the way.
What Meta Says
Stone's rebuttal centers on the idea that the permissions architecture is designed to keep users in the driver's seat. By requiring layered consent, Meta argues, the system builds in safeguards against the kind of unauthorized access the journalist described.
The incident highlights the growing tension around AI agents that integrate deeply with personal data to deliver more tailored responses. As these tools expand their reach across messaging, email, and other sensitive areas, questions about consent and transparency are becoming central to the conversation.
For users weighing whether to adopt such assistants, the episode underscores a few key considerations:
- How permissions are structured and whether they can be revoked
- What data an AI agent can access by default
- Whether company explanations match users' real-world experiences
The clash between Meta's official position and the journalist's firsthand account leaves open questions about whether the discrepancy stems from a misconfiguration, a misunderstanding of the settings, or something else entirely. For now, Meta is standing firmly behind its claim that consent remains in the hands of the user.