Meta has halted a sharing feature in its Muse AI assistant following the revelation that thousands of shared conversations were accessible online. Security researchers from the cybersecurity firm Claroc uncovered this issue, which involved over 6,800 web addresses linked to shared conversations. Many of these addresses contained sensitive information, including confidential business documents and personal financial details.

The findings were reported by Ars Technica, highlighting that researchers could access numerous pages without needing to log into a Meta account. Among the exposed materials were a marketing company’s internal revenue review, a small business owner’s loan application, a financial report from a real estate investor, and a workplace calendar detailing employees’ planned vacations. Additionally, notes for a confidential call with a senior figure at an AI lab were also found.

In response to these findings, Meta has paused the sharing feature while it assesses the situation and considers modifications to prevent such vulnerabilities in the future. Tracy Clayton, a spokesperson for Muse, confirmed the suspension and stated that the ability to share specific Muse artefacts created accessible pages for those with the link.

How the Exposure Occurred

Muse AI allows users to share conversations, images, and other content via web links. This functionality is intended to facilitate the sharing of AI-generated material among colleagues or other individuals. However, Claroc identified a flaw in how Muse generated these links, which included a brief description of the conversation followed by a random identification code. This design meant that the web addresses could disclose the content of the shared pages even before they were accessed.

Researchers employed passive DNS records, which retain information about observed domain names, to locate these addresses. Once a shared page was accessed, its web address could be logged in searchable records, allowing researchers to reconstruct the link and retrieve the conversation. Claroc indicated that identifying over 6,800 such addresses required merely a single search query, and the researchers examined only a fraction of the links discovered.

The findings were reported to Meta’s bug bounty programme on October 8. Asaf Aprozper, co-founder and chief technology officer of Claroc, also attempted to reach out to Meta representatives through LinkedIn to discuss the issue further.

Overview of Muse AI

Muse AI was launched on September 8 as a versatile AI assistant designed to perform various tasks, including filling out forms, booking appointments, making purchases, and creating documents. These functionalities often require access to users’ accounts and personal information, raising concerns about data security.

The recent discovery has intensified scrutiny regarding the protection of shared AI conversations, especially when users discuss sensitive topics such as financial matters, workplace information, or confidential business strategies. Claroc noted that it found no evidence of misuse of the exposed conversations but emphasised that its findings represented only one method of uncovering shared pages and did not fully capture the extent of the exposure.

This incident follows previous occurrences where shared conversations from AI assistants, including ChatGPT and Grok, were found in Google Search results, highlighting ongoing concerns about data privacy in AI technologies.

Long or Short, get news the way you like. No ads. No redirections. Download Newspin and Stay Alert, The CSR Journal Mobile app, for fast, crisp, clean updates!

App Store – https://apps.apple.com/in/app/newspin/id6746449540

Google Play Store – https://play.google.com/store/apps/details?id=com.inventifweb.newspin&pcampaignid=web_share