Meta Pauses Muse Sharing After Privacy Leak

Overview

Meta has suspended the Muse sharing feature following a significant privacy breach that exposed sensitive user data to the public internet. The pause comes after security researchers at the firm Claroc discovered thousands of shared AI conversation links that were easily discoverable through public DNS records.

The incident highlights growing concerns about how major technology companies handle user privacy when integrating sharing capabilities into AI assistants. Meta confirmed in a statement that the functionality allowing users to share specific Muse artifacts is currently inactive while the company reviews the findings and considers changes to link discovery mechanisms.

The Privacy Breach

Researchers at Claroc published a report this week detailing how they identified more than 6,800 publicly accessible Muse links. These links, intended for private sharing between users, were exposed because of a flaw in how Meta generated their web addresses.

According to Claroc, each shared conversation page was assigned a URL composed of the share’s title, a short descriptive phrase, and a random identifier. This structure inadvertently made the links predictable. Because the hostname often revealed the content of the page before it was even opened, investigators could reconstruct valid URLs by pulling hostnames from public DNS records.

“People have the ability to share specific Muse artifacts, which creates a page that can be accessed by those with the link,” Muse spokesperson Tracy Clayton said in an emailed statement to The Verge. “However, we’ve paused this functionality while we review this report and consider changes to how these links can be discovered.”

The consequences of this technical oversight were immediate and severe. By exploiting the predictable nature of these URLs, Claroc was able to view hundreds of conversations that contained highly personal and proprietary information.

Exposed Sensitive Data

The scope of the data exposure went far beyond casual chat logs. The shared links revealed documents and discussions that users reasonably expected to remain private. Among the findings were:

• Financial records: A marketing company’s internal revenue review was visible to anyone who found the corresponding link.
• Personal financial applications: A small business owner’s loan application, containing sensitive financial history, was publicly accessible.
• Workplace calendars: A full-year staff calendar for a workplace, detailing schedules and potentially private meetings, was exposed.
• Professional communications: Notes prepared for a job candidate’s call with an AI lab were also discoverable through the flawed links.

This breach illustrates the real-world risks associated with AI sharing features. When users share a conversation, they often include context that they assume is only visible to the recipient. The transparency of these links undermined that assumption entirely, turning private dialogues into publicly searchable content.

A Pattern in AI Security

This incident is not isolated within the artificial intelligence sector. Similar vulnerabilities have plagued other major AI platforms, suggesting a systemic challenge in balancing shareability with security.

Previously, conversations with OpenAI’s ChatGPT, Anthropic’s Claude, and xAI’s Grok have been found indexed on Google Search or otherwise exposed on the public web. In those cases, users who shared links often forgot that the default settings allowed for broader access than intended. The Meta incident, however, was notable for the ease with which thousands of links could be discovered programmatically via DNS records rather than through manual searching.

Security experts warn that as AI assistants become more integrated into professional and personal workflows, the volume of sensitive data shared via these platforms will increase. Features designed to facilitate collaboration must be robust enough to prevent accidental exposure, especially when the data involved includes corporate secrets, financial applications, and personal schedules.

Meta’s Response and Next Steps

In response to the Claroc report, Meta moved quickly to disable the vulnerable feature. The company stated that it is conducting a thorough review of the report and is evaluating how to redesign the link discovery process to prevent similar exposures in the future.

The immediate suspension of the sharing capability affects all users of the Muse assistant. Until Meta implements more secure sharing protocols, such as requiring authentication for link access or using non-predictable, encrypted identifiers, users will not be able to share their AI conversations externally.

This event serves as a stark reminder for both technology providers and consumers. For providers, it underscores the need for rigorous security testing of sharing features before deployment. For users, it highlights the importance of verifying privacy settings and understanding the potential reach of shared digital content. As AI tools continue to evolve, maintaining trust through robust data protection will remain a critical priority for companies like Meta.