Meta has temporarily disabled the sharing feature for its Muse AI assistant. The decision follows a report from security firm Claroc, which discovered that thousands of private user conversations were accessible via public web links.
Researchers at Claroc identified over 6,800 Muse links by analyzing public DNS records. The firm noted that the structure of the URLs, which included descriptive titles and IDs, made them predictable and easy to locate through directory searches.
The exposed data included sensitive information, such as revenue reviews for marketing companies, loan applications from small business owners, internal staff calendars, and notes from job interviews. Meta is currently reviewing the report to determine how to improve link security and discovery.
This incident adds to a growing trend of AI platforms inadvertently exposing user data. Similar issues regarding the indexing of private chat logs have previously affected services from OpenAI, Anthropic, and xAI.
