Meta Restores Muse AI Sharing After Fixing Public Link Exposure Flaw
Meta has switched sharing back on for its Muse AI tool after patching a vulnerability that allowed strangers to view sensitive documents via predictable web addresses.
Key points
- Meta re-enabled Muse sharing after fixing an issue where public links were easily discoverable.
- Security researchers from Claroc demonstrated access to internal business documents using these exposed URLs.
- The vulnerability affected shared artifacts like documents and images, not the core chat functionality.
Meta has resumed the sharing feature for its Muse AI assistant after addressing a security flaw that allowed unauthorised access to private content. The company confirmed that the underlying issue is resolved and that the service is now operational again for users who wish to share their interactions.
Security researchers had previously warned that public conversation links were too easy to locate. Meta paused the sharing capability following reports from Claroc, a security firm, which demonstrated how attackers could find and view sensitive information hosted on these publicly accessible web pages.
View the original post on X
What we know
Researchers at Claroc reported earlier this week that the web addresses for shared Muse conversations were named after the conversation titles and written into public DNS records. This structure made it simple for anyone to guess or find specific links without needing an invitation or password.
Meta’s David Singleton stated on X that the fix specifically targeted the URLs for shared Muse artifacts, which include documents and images. He clarified that the flaw was not present in the general chat sharing mechanism, limiting the scope of the exposure to these specific file types.
Singleton explained that any artifact shared before the fix now possesses a new URL. Visitors attempting to access the old links will see a message requesting the new address from the original sender. Meta apologised for the disruption but emphasised that protecting user data was the priority.
According to The Verge, researchers had successfully accessed pages containing a marketing company’s internal revenue review and a small business owner’s loan application. These examples highlighted the potential severity of the exposure before Meta intervened to disable the sharing feature temporarily.
Ars Technica reported that Meta paused the sharing feature after being contacted about the issue. The company acted quickly to mitigate the risk, ensuring that no further unauthorised access could occur through the vulnerable URL structure during the patching process.
Defender's checklist
- Review all previously shared Muse artifacts to ensure they contain no sensitive or confidential information.
- Notify any external parties who received old links that they must request new, secure URLs from you.
- Audit your organisation’s current sharing policies to ensure AI-generated content is handled securely.
- Monitor public DNS records for any unexpected entries related to your company’s AI interactions.
The background in plain language
When organisations use AI tools, they often generate documents or images that require sharing. If these files are hosted on public web addresses that are easy to guess, anyone on the internet can view them. Teams usually need to ensure that shared links are unique and unguessable. Check whether your AI tools generate predictable URLs. Ensure that sensitive data is never placed in publicly accessible locations without strong access controls.
What to do and how to stay safe: Meta
- Identify any Muse artifacts you have shared recently and verify if they contain sensitive business or personal data.
- Contact recipients of old links to inform them that the links are invalid and request new ones if further access is needed.
- Review your organisation’s guidelines on sharing AI-generated content to prevent accidental exposure of confidential information in the future.
- Monitor for any reports of unauthorised access to your shared AI content and report such incidents to your IT security team immediately.
General security guidance from the TechWire Beat newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Did Meta fix the vulnerability in Muse AI?
Yes, Meta confirmed that the underlying issue with shared artifact URLs is now fixed and sharing is back on.
Was the chat sharing feature affected by this flaw?
No, Meta stated that the issue was with URLs for shared artifacts like documents and images, not with chat sharing.
What happens if someone tries to use an old shared link?
Visitors to the old link will see a note asking them to get the new URL from the original sender.
