Muse, the new general-purpose AI agent from Meta, can expose sensitive information when users share conversations through Web links, researchers reported Monday.
Like many online services, Muse allows users to share searches, conversations, images, and similar content with other people. Such share features usually involve clicking an option to create a unique URL that anyone can use to load a page hosting the content. To keep potentially sensitive information from being accessible to unintended people, the URLs typically contain a long sequence of text that can’t be easily guessed.
A single query is all that’s required
Muse, which Meta CEO Mark Zuckerberg said last month was “built from the ground up for privacy and security,” seems to take the same precautions. But a central design decision for constructing URLs has made the links easy to discover. As a result, researchers from security firm Claroc said they were able to enumerate 6,800 links Muse users created for sharing conversations with others. After opening a few hundred of them, the researchers found information that was clearly not intended to be publicly available, including:
- a candidate’s notes preparing for a confidential executive-search call with a frontier AI lab, naming the lab’s internal contact and the meeting time
- a national marketing company’s internal revenue review, marked for its CEO
- a small-business owner’s loan application, with the amount requested
- a to-do list that included a bank fraud alert, with a partial account number and a transaction amount and time
- a real-estate investor’s internal performance report, with two property addresses and exact return, income, and mortgage figures
- a workplace’s full-year staff vacation calendar, listing dozens of employees against the week each one would be away
“If you had asked us beforehand, we would have guessed that gathering these would take real effort,” the researchers wrote. “It took one query.”
The easiest way to discover the URLs is to perform a passive DNS search through one of many available commercial services. Passive DNS records store the historical resolution of domain names to IP addresses, as well as nameserver, and mail-server records over time. Passive DNS systems typically collect DNS responses from sensors positioned at recursive DNS resolvers or other infrastructure. A passive DNS record is generally created when a monitoring system observes a DNS response. In the case here, once the intended party clicks on the share link, the DNS information is recorded in a searchable database.
Claroc researchers were able to discover the Muse share links because all of them use the domain name cf.metaaiusercontent.com. Discovery is further made easy by the decision to add the conversation name to the URL, along with a long string of text. Simplified examples of such links are home-sale-budget-9f21c8a3.cf.metaaiusercontent.com, q3-revenue-review-7b1e42f0.cf.metaaiusercontent.com, and loan-application-draft-a4c9e2.cf.metaaiusercontent.com. It’s this last design to include the conversation title in the URL decision that makes the public discovery possible.