New turn in Meta Muse privacy case as share links leak chats

In October 2026, the president of the Signal Foundation told that Muse puts the data of contacts at risk. Meta told that Muse uses public information and the data that the user gives to it.

Earlier: Signal president says Muse puts the data of contacts at risk

The security firm Claroc found that the share function of the Meta Muse AI agent shows personal chats. Ars Technica tells of the result, and Meta told Ars that it stopped the function. Mark Zuckerberg had told that Muse is built from the ground up for privacy and security, and a community note on X corrected the post.

How it works

A share link has the title of the chat and a random string of characters. When a person opens a link, the web address goes into a lookup dataset that all can read. The researchers looked in that dataset for other Muse share addresses.

What they found

Claroc opened pages from more than 6,800 links. The chats had corporate revenue documents, loan applications, bank transaction details, income details and a workplace vacation calendar. The researchers tell that it needed one lookup.

What is not known

The source does not tell how long the links were open. It does not tell how many users the leak touched. Meta gave no other comment.

Insight: Wide access makes each defect cost more

Muse needs the user to give it access to bank accounts, passwords, contacts and chat logs. Meta seems to push Muse out fast, and a small defect in the share function shows data that the agent holds on purpose. Leaders must set a rule for which data staff can give to a consumer agent.