Meta’s Muse AI assistant isn’t limited to just answering questions or making everyday tasks easier. It can also create detailed profiles of users’ preferences, habits and social relationships based on information related to their chats, emails and Instagram. A Time report analyzed Muse’s internal instructions, raising new questions about AI privacy and personal data.
Muse is designed to help with tasks like buying groceries and canceling subscriptions. But for this it can also continuously collect and analyze information about the user.
User’s profile is updated every hour
According to TIME’s report, Muse updates its user’s profile every hour. For this, it can take information from chats, messages and emails to which it has got access. This system is not limited to just the user’s likes and interests, but also tries to understand his relationships with other people.
This profile may include information about how the user met the person, what the two like, whether they have ever had disagreements, or whether there has been tension or mutual support in their relationship.
According to the report, Muse can also guess those goals of the user, which he did not tell directly. This system also tries to understand which type of messages or reminders can have the most impact on the user. For example, its internal instructions mention suggestions like sending short messages after 10 pm. Apart from this, AI analyzes the conversations throughout the day even at night.
Another aspect of this is that information about people mentioned in conversations can also become part of profiling, even if they have not used Muse themselves.
What clarification did Meta give?
Meta did not reject the information revealed in TIME’s report. The company says that Muse remembers the information that users themselves share, so that it can provide more useful answers and help in the future.
According to the report, Muse has been downloaded more than 50 lakh times so far and after its release it reached the number-1 position on America’s App Store. Meta runs each Muse Assistant on a separate virtual machine in its cloud infrastructure. Users can also export copies of their Muse files.
The company claims that this user data is not shared with its advertising system or other Muse agents. However, some information gained from AI experiences can be used to improve the product. This also includes understanding what types of reminders are more effective. According to Meta, identifying details are removed before such information can be used.
What happens if I delete information from Muse?
Meta says that users can ask Muse to forget certain information. But TIME’s investigation revealed that asking someone to forget information does not necessarily delete the original message associated with it. According to the report, Muse’s internal instructions also include not prominently informing users about this difference.
Meta has announced plans to introduce such encryption by the end of this year, due to which even the company will not be able to access Muse’s data. Currently this security is not available. Users can currently view their files and session logs through Muse’s file browser.
After this report, the debate on the balance between convenience and privacy of AI assistant may intensify. The more personal context AI remembers, the more important it will be to have clarity about what information is saved, how long it lasts, and how it can be completely deleted.
