Meta’s Muse AI Agent Leaks Private User Data

When Meta rolled out its much‑hyped Muse AI assistant last month, the company’s CEO, Mark Zuckerberg, took to X (formerly Twitter) to assure users that the chatbot was "built from the ground up for privacy and security." It was a comforting promise—until a single query turned that promise on its head.

According to Ars Technica, a team of security researchers from the firm Claroc discovered a surprisingly easy way to pull up private content that Muse had supposedly safeguarded. The trick? Muse’s built‑in "share" button, which creates a unique web link whenever a user wants to send a chat transcript or any other piece of information to a friend.

The generated URL looks innocuous enough: a short, randomly‑generated string appended to a generic Muse address, plus the AI‑crafted title of the conversation. But behind the scenes, every time such a link is opened, the address is logged in a massive lookup table—a kind of phone‑book for URLs that have been visited.

Claroc’s researchers queried that lookup table, filtered out the Muse share links, and—boom—found thousands of other URLs that pointed to completely unrelated chat logs. Some of those logs contained corporate revenue spreadsheets, loan applications, bank transaction histories, personal income statements, and even a workplace vacation calendar. The data was laid out in plain text, ready for anyone with the link to read.

“If you had asked us beforehand, we would have guessed that gathering these would take real effort,” the team wrote. “Instead, it took one query.” In other words, a single, well‑crafted request unlocked over 6,800 private pages.

Meta quickly told Ars that the offending feature had been disabled, but the incident is far from an isolated slip‑up. Earlier in the month, other reports surfaced showing that Muse was automatically generating a “page” for every person in a user’s life—friends, family, coworkers—essentially building a dossier on people the user never explicitly asked the AI to track.

All of this adds up to a worrying pattern: an AI that markets itself as privacy‑first but silently amasses, stores, and sometimes inadvertently shares intimate details about its users. If Muse’s first few weeks in the wild are any indication, we may be witnessing just the tip of the iceberg for this seemingly cute chatbot.

For now, users are left wondering whether they can truly trust an AI that promises secrecy while quietly keeping a detailed ledger of their lives. As the conversation around AI ethics and data protection continues to evolve, Muse serves as a stark reminder that good intentions aren’t enough—robust, transparent security measures are essential.