Meta Halts Muse Chat Sharing After 6,800 Links Surface in DNS Records

The News

Meta has switched off the link-sharing function in Muse, its AI assistant, after a security firm showed that supposedly private shared conversations could be located through ordinary internet infrastructure. Researchers at Claroc said they identified more than 6,800 Muse links by trawling public DNS records, the directory system that routes web traffic.

The flaw lay in how each shared page was addressed. According to Claroc, every link combined a short title describing the conversation with a random ID, and that same string surfaced in the hostname. Anyone reading DNS data could therefore guess what a page contained and reconstruct the full link without being sent it.

Using this method, the firm said it viewed hundreds of conversations. Among them were a marketing company's revenue review, a small business owner's loan application, a full-year staff calendar for one workplace and preparation notes for a job candidate's call with an AI lab.

A spokesperson, Tracy Clayton, told The Verge that shared Muse artifacts create a page reachable by anyone holding the link, and that the feature has been paused "while we review this report" and weigh changes to how such links can be discovered.

Why It Matters

This is the latest in a string of episodes where a convenience feature quietly widened the audience for chatbot conversations. Shared chats from OpenAI's ChatGPT, Anthropic's Claude and xAI's Grok have all previously turned up in Google Search results. In mid-2025 OpenAI withdrew an option that let shared ChatGPT conversations be indexed by search engines after users found personal material exposed.

The Muse case differs in one important respect: users did not need to tick a "make discoverable" box. The exposure came from link design, which places the burden on the platform rather than the individual. For companies, that is the uncomfortable lesson. Staff are pasting revenue figures, calendars and credit paperwork into consumer assistants, and the security of that data depends on engineering choices they never see.

Indian Angle

India's Digital Personal Data Protection Act, 2023, and the DPDP Rules notified in November 2025, treat a platform that decides how personal data is processed as a data fiduciary responsible for "reasonable security safeguards". Once the breach-notification provisions take full effect, an exposure of this kind involving Indian users would need to be reported to the Data Protection Board and to affected individuals. Separately, CERT-In's 2022 directions already require reporting of specified cyber incidents within six hours.

The loan-application example should concern Indian lenders and fintechs in particular. Many MSME owners now draft credit proposals, GST summaries and bank-statement analyses with free chatbots before approaching banks or NBFCs. RBI's digital lending framework governs regulated entities and their lending service providers, but it does not reach a borrower's own use of a consumer AI tool, which leaves that data outside any lender's control.

For Indian enterprises, from IT services majors to D2C start-ups, the practical response is a written policy on which assistants may handle client or financial data, with enterprise tiers that disable public sharing by default. Domestic model builders such as Sarvam AI and Krutrim, which pitch data residency, may gain ground with compliance teams.

FAQ

What exactly was exposed?

Claroc said it found more than 6,800 Muse share links via public DNS records and viewed hundreds of conversations. Examples included a company revenue review, a loan application, a workplace staff calendar and interview preparation notes. Only conversations that users had chosen to share through a link were affected, not every private chat.

Has Meta fixed the problem?

Meta has paused the sharing feature while it reviews the research and considers changes to how links can be discovered. It has not, according to the report, announced a permanent redesign or a timeline for restoring the function.

What should Indian users and businesses do now?

Avoid placing account numbers, PAN or Aadhaar details, loan documents or client data into consumer chatbots. Review any conversations you have shared via link and delete them where the option exists. Businesses should adopt an approved-tools list and prefer enterprise plans with sharing controls and clear data-processing terms that align with DPDP obligations.

Where can I read the original report?

The Verge covered the pause and Meta's statement, citing earlier reporting by Ars Technica and the research published by Claroc. The full source link appears below.

This story was reported by The Verge. Read the full original coverage at The Verge.