Meta* suspended sharing Muse chats after data leak via DNS
Over 6,800 public DNS links exposed private Muse chats, revealing financial reports, staff calendars and credit applications; Meta* halted sharing, highlighting industry risk.
How the leak occurred
Muse’s sharing feature creates a web page accessible via a direct link. As Meta* representative Tracy Clayton explained, that was the intended design: the user shares a result and the recipient opens the page. The problem lies in how these pages are registered in the domain name system. According to The Verge, citing Claroc’s report, the page address contained the chat title — for example, “Q3 revenue overview” or “sales department calendar” — plus a unique identifier. DNS records stored these hostnames in plain text.
Researchers didn’t need to hack accounts or guess passwords; they only had to request public DNS data, filter records belonging to Muse subdomains, and collect ready-made links. Because the chat title was part of the address, an attacker could see what was inside without opening the page. A total of 6,800 links were found. The publication does not specify how many led to truly sensitive data, but the examples speak for themselves.
What data was exposed
The list of Claroc-found dialogues reads like a company dossier. Exposed were the marketing firm’s revenue overview — figures, margins, revenue structure; a full year-long staff calendar for one workplace — vacations, shift changes, meetings; a credit application for a small business with financial metrics and the owner’s personal data; and interview notes for an AI-lab candidate — competency assessments, salary expectations, internal HR comments.
Meta* response and systemic market issue
Meta* suspended the sharing function while it reviews the report and decides how to make links less discoverable. In a statement to The Verge, Tracy Clayton called this precautionary measures. The fix will likely affect link generation — removing the title from the URL, using longer tokens, and restricting DNS zone enumeration. However, the root cause is not Muse’s code but the industry approach: collaboration features are released faster than threat detection is addressed.
What this means for small businesses in Russia and the CIS
Russian companies are actively integrating AI into marketing, sales, accounting, and HR. Employees upload revenue, margins, client databases, contract drafts, and payroll reports into chats. This is often done in personal accounts of free versions, without corporate policies. The “share” function is used to quickly send results to a colleague via Telegram or email with a single click.
Five steps to take this week
- First, conduct an audit of the AI tools used. List services where employees work with business data — chatbots, text generators, spreadsheet analyzers. Identify which have a link-sharing feature and whether it is enabled by default.
- Second, prohibit sharing links to chats containing work data until privacy settings are verified. If a service cannot disable public links or guarantees they are not indexed by DNS or search engines, do not use it for confidential information. Send the result by copying text into a secure corporate messenger or email.
- Third, establish a rule to cleanse entered data. No real names, tax IDs, account numbers, salaries, or client addresses in queries. Replace them with placeholders such as “Client_1”, “Contract_Amount”, “Region_South”. AI analysis doesn’t need real identifiers; it only needs structure and numbers.
- Fourth, set up DLP control or at least an information security policy that blocks or logs access to domains like .meta.ai, chatgpt, claude.ai and similar from corporate devices, unless there is a provider agreement. For small businesses without an IT security team, router or antivirus DNS filtering is sufficient.
- Fifth, record the policy in a document and give the team a brief briefing. Explain: “The ‘Share’ link is not a private file; it’s a potentially public web page.” Use the Muse example. People understand risks better through concrete stories.
Summary
Convenience of sharing features in AI tools outpaced security. Meta* halted Muse, but the architectural issue is universal: a public link is a perimeter hole that scanners find faster than you can close it. For business, the takeaway is clear: treat any AI chat as a potentially public document. Don’t upload secrets, disable link sharing, and cleanse entered data. This isn’t paranoia — it’s data hygiene for 2024.
Frequently Asked Questions
What is a DNS leak?
A DNS leak is a situation where confidential information becomes accessible through the domain name system. In the case of Muse, links to private chats were discovered in public DNS records.
Which data could have been compromised in the Muse incident?
According to the report, links to private chats containing financial information, calendar entries, and personnel notes were compromised. This indicates a leak of sensitive business data.
How can similar leaks be prevented in the future?
To prevent such leaks, you should use secure data exchange protocols, enforce strict access controls, and conduct regular security audits. It is also important to deploy data loss prevention (DLP) systems.
- Meta Platforms Inc. (social networks Facebook and Instagram*) is recognized as extremist and banned in the Russian Federation.
- Meta Platforms Inc. (including social networks Facebook and Instagram*) is recognized as an extremist organization and its activities are prohibited in the Russian Federation.
Web development, digital advertising and marketing expert since 2012. Founder of AIFORSALE.