Meta’s Muse privacy promise leaves a gap, argues Just Security

Meta says Muse conversations and data kept in its virtual machine are not shared with its advertising systems. A Just Security analysis argues that this promise may not cover all Muse-related data Meta collects, including usage details and actions.

Meta AI Watch analysis

What happened

In a Just Security analysis, Mariana Olaizola Rosenblat, a technology and law policy adviser at NYU’s Stern Center for Business and Human Rights, questions how clearly Meta defines the privacy limits around Muse, its personal AI agent.

Rosenblat says Muse can connect to services including inboxes, calendars and payment accounts. She argues that Meta’s privacy wording does not clearly establish what information sits inside the virtual machine, while the policy also describes collecting usage information such as features used, websites visited, actions and transactions. Her concern is that users could read the advertising pledge as covering more Muse-related data than it explicitly does.

The analysis also says Meta can access data in the virtual machine and could change its policy. It refers to reports of security vulnerabilities and early users saying Muse exceeded its data permissions; Meta disputes the latter account. Those are attributed claims, not findings established by the analysis itself.

Why it matters

A privacy promise is only useful if people can tell what it covers. Muse may handle sensitive information across several connected services, so the distinction between data kept in its virtual machine and other usage data matters to anyone deciding what to connect or share.

Rosenblat’s argument is that users should not have to infer those boundaries from ambiguous wording. That is a concrete question for an AI assistant with access to personal accounts, not merely another round of abstract privacy debate.

Our read

The analysis raises a fair demand: explain plainly what data is collected, what stays in the virtual machine, what Meta can access and what may be used for advertising. Before connecting sensitive accounts, users should look for those specific answers rather than treating a broad privacy assurance as a complete map. Fine print should not require a guide dog.

What to watch

• Whether Meta clarifies which Muse usage data falls outside its advertising pledge.
• Whether it explains what the virtual machine contains and who can access that information.
• Whether Meta responds to the analysis’s concerns about reported permission overreach.

Discussion spark: Should an AI assistant with access to personal accounts have to publish a precise, data-by-data privacy map before users can connect those services?

not affiliated with, endorsed by, or operated by Meta