Meta suspends Muse chat sharing after private details were exposed
Meta has temporarily disabled sharing of conversations with its Muse AI agent after security researchers found a way to access thousands of them. The company is reviewing how the links can be discovered.
Meta, the technology company behind Facebook and Instagram, has suspended the chat-sharing feature of its Muse artificial intelligence platform after researchers at security firm Claroc found that anyone could potentially find and open users' private conversations. Muse is an agentic AI platform, meaning it can carry out tasks on a user's behalf, and Meta introduced it last month. The privacy problem surfaced shortly after launch.
How the links could be found
To share a conversation, a user created a URL. Claroc explained that all of these links used the same domain name and put the conversation's title in the subdomain. That address was recorded in public DNS data. Passive DNS consists of records of how human-readable domain names resolve to the IP addresses browsers use, essentially working like a phone book for the internet.
In a blog post, Claroc said a single passive-DNS search returns thousands of such addresses, listed by name like a table of contents.
What was exposed
The researchers enumerated 6,800 links to shared Muse chats and were able to view private information from both business and personal uses of the service. The exposed material included meeting times with a frontier AI lab, a small-business loan application, a bank fraud alert showing partial account details, a company's vacation calendar and other internal company documents.
Meta's response
The findings were first reported by Ars Technica. Meta told that publication it had paused the functionality while it reviews the report and considers changes to how the links can be discovered. Claroc also noted that, after Meta's update, its researchers could not open a sample of the links they had accessed earlier.
