Meta suspends Muse chat-sharing feature after more than 6800 sharing links were discovered

AT A GLANCE

It was reported that Meta Muse chat-sharing links could be retrieved in bulk through DNS historical records, after which Meta suspended the sharing feature.

Article

Meta Muse Privacy vulnerability exposed: a single search found 6800 multiple chat links, even corporate secrets were uncovered

Beating AI Flash: Claroc, a U.S. cybersecurity company, found that Meta personal AI Agent Muse chat-sharing links could be retrieved in bulk by strangers. Researchers queried a DNS historical database just once and found more than 6800 shared links. After spot-checking several hundred, they found dozens of sensitive materials, including internal corporate revenue reports, small-business loan applications, and preparation materials for secret recruiting communications between a candidate and a frontier AI lab; even internal contacts and meeting times were exposed.

When users ask Muse to share a chat, the system generates a webpage that can be viewed without logging in and writes the chat title and a random ID into the domain name. Once someone opens the page, the full domain name may be recorded by third-party DNS services. Claroc directly searched these historical records and obtained sharing addresses that would otherwise have been difficult to guess. Although Muse asks search engines not to crawl these pages, it cannot prevent the domain names from being recorded by other services.

Claroc reported the vulnerability to Meta as early as October 8, and October 18 published its research. It was not until October 20, after U.S. technology media outlet Ars Technica intervened with inquiries, that Meta confirmed it had suspended the chat-sharing feature. Claroc rechecked 40 links that had previously been accessible without logging in that day and found that all had become invalid, but chat titles remained in DNS historical records.

Meta CEO Zuckerberg claimed as recently as last month that Muse was a product “designed for privacy and security from the ground up.”

Original link https://m.theblockbeats.info/flash/372971

Key points

Beating AI reported that Claroc, a U.S. cybersecurity company, found that Meta personal AI Agent Muse chat-sharing links could be retrieved in bulk by strangers.
It was reported that, after researchers queried a DNS historical database once, they found more than 6800 sharing links.
It was reported that, after spot-checking several hundred links, Claroc found dozens of sensitive materials, including internal corporate revenue reports, small-business loan applications, and preparation materials for secret recruiting communications.
When users ask Muse to share a chat, the system generates a webpage that can be viewed without logging in and writes the chat title and a random ID into the domain name.
It was reported that, after a page is opened, the full domain name may be recorded by third-party DNS services, and Claroc obtained sharing addresses by searching historical records.
It was reported that Claroc reported the vulnerability to Meta in October 8 and published its research in October 18.
It was reported that Meta confirmed in October 20 that it had suspended the chat-sharing feature.
Claroc rechecked 40 links that had previously been accessible without logging in that day and found that all had become invalid, but chat titles remained in DNS historical records.
The article said that Meta CEO Zuckerberg claimed last month that Muse was a product “designed for privacy and security from the ground up.”

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

Muse’s sharing feature generates public webpages. The issue described in the report was not that a random ID was guessed directly, but that the full domain name containing the chat title and a random ID could be stored by third-party DNS services, allowing strangers to find sharing addresses that would otherwise have been difficult to guess from historical records.

Why it matters to readers

The report claimed that the retrieved links contained sensitive materials such as corporate revenue, loan applications, recruiting communications, internal contacts, and meeting times; even if webpage links became invalid, chat titles could still remain in DNS historical records.

When using chat-sharing links that do not require login, note that the links and their titles may be recorded by third-party services; the article states that Meta has suspended this sharing feature.

Risks and unknowns

The only evidence is Dongcha Beating's AI retelling; the original Claroc research, Meta statement, and Ars Technica report were not provided, so the relevant claims cannot be independently verified.
More than 6800 links, dozens of sensitive materials, and the accuracy of their specific contents cannot be confirmed from the available evidence.
The total scope of affected users, the duration of link exposure, and Meta's subsequent notification measures are not specified.
40 Apart from invalid link, it is unclear whether all other shared links have also become invalid.

Meta suspends Muse chat-sharing feature after more than 6800 sharing links were discovered

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center