Meta's AI Assistant Muse Exposed Sensitive Data Through Security Flaw
A security firm found over 6,800 links generated by Muse were publicly accessible, revealing private user information.
Meta's AI assistant, Muse, has a security vulnerability that allowed sensitive personal and business information to be exposed, according to a report by security firm Claroc.
The investigation discovered a flaw in Muse's sharing feature, which led to over 6,800 links created by the AI becoming publicly accessible. These links contained user data that was intended to be private.
According to Ars Technica, the exposed information included details such as loan application amounts, bank transaction data, a real estate investor's internal performance reports, and corporate revenue documents. One instance revealed a company-wide vacation calendar detailing employees' time-off dates. The Claroc researchers noted that accessing this sensitive data required minimal effort, stating, "It took one query."
The vulnerability was exploited through Muse's share function, which allows users to share searches, conversations, and other content. While these shared items were intended to be accessed via AI-generated web links with long, cryptic character strings, Claroc found that Muse's URLs were easily identifiable. They all used the domain name cf.metaaiusercontent.com and included the AI-generated conversation title alongside the text sequence, making them discoverable without the intended recipient being present.
Claroc referred to these instances as "GhostShares," describing them as "pages that open for people who were never handed the link, and that those people can find on their own."
In response to the findings, Meta has temporarily disabled the sharing functionality. "We’ve paused this functionality while we review this report and consider changes to how these links can be discovered," a company representative told Ars Technica.
This incident follows other privacy concerns raised regarding Meta's technology. Last week, users reported that Meta Quest smart glasses appeared to be recording audio even when deactivated. Meta AI has previously faced scrutiny, including user comments about its mascot, Jolly, being designed with a "thirst trap body."