The decision came after researchers at cybersecurity firm Clarok identified more than 6,800 shared chat links through passive Domain Name System (DNS) records, raising concerns about the privacy of users who had shared conversations through the AI platform.
According to the researchers, several hundred conversations were accessible through the identified links. The exposed material reportedly included corporate revenue assessments, loan applications, bank fraud alerts, partial account details and confidential notes relating to job interviews.
The findings highlighted a vulnerability in the way Muse generated and handled shared conversation links, potentially allowing individuals who had not received the links directly to discover and access the associated chats.
How the privacy issue emerged
Researchers attributed the problem to Muse’s URL structure, which incorporated conversation titles alongside randomly generated identification codes.
When users accessed these shared links, the web addresses could become visible in publicly available DNS-related records. This created a potential route for third parties to discover links that users may have intended to share only with selected individuals.
The researchers used passive DNS records, which collect historical information about domain-related activity, to identify thousands of such links.
They reported that access to the conversations did not require users to sign in to a Meta account, further increasing concerns about the exposure of information shared through the platform.
Although the links were generated through a sharing feature, their discoverability through external records meant that conversations could potentially reach a wider audience than users had anticipated.
Sensitive business and personal information at risk
The investigation found that some accessible conversations contained information of a sensitive financial and professional nature.
These included business revenue reviews, details associated with loan applications, warnings about suspected banking fraud and confidential interview preparation notes.
The findings raised concerns about the handling of information submitted to AI assistants, particularly as individuals and businesses increasingly use such platforms for financial analysis, professional communication and other work involving confidential material.
The reported vulnerability also highlighted the distinction between deliberately sharing a conversation through a generated link and making that link discoverable to individuals outside its intended audience.
Meta disables sharing feature
Following the disclosure, Meta suspended Muse’s chat-sharing functionality.
The move was intended to address concerns surrounding the accessibility of shared conversations and the possibility of sensitive information being discovered through publicly available records.
The reported incident underscores broader data protection challenges facing AI platforms as their use expands across personal and professional activities.
It also highlights the importance of secure link generation, appropriate access controls and safeguards designed to prevent confidential conversations from becoming accessible beyond their intended recipients.
For users and businesses relying on AI assistants, the findings serve as a reminder that shared conversations may contain information requiring the same level of protection as other sensitive digital records.