Muse, the latest AI assistant from Meta, has been found to expose sensitive information when users share conversations through web links, despite being designed for privacy. Researchers identified 6,800 such links, revealing candid executive notes, financial reports, and personal schedules. Meta has since suspended the feature for review.

Typically, sharing features on web services involve unique, lengthy URLs that are hard to guess. However, Muse's design flaw made it easy for researchers to discover and access these links. Examples included a candidate's executive-search call notes, a marketing company’s internal reviews, and a small business’s loan applications. The ease of accessing these private details highlights the need for robust privacy measures in AI technology.

The findings came as a surprise, given Meta CEO Mark Zuckerberg's claim that Muse was built with privacy in mind. Asaf Aprozper, co-founder of Claroc, reported the issues through Meta's bug bounty program but received no response. Meta acknowledged the issue and paused the feature, promising to consider changes to prevent such leaks in the future.

This incident underscores the ongoing challenge of balancing convenience and privacy in digital services. As AI assistants become more integrated into daily life, ensuring user data remains secure will be crucial for maintaining trust.