--- title: "Meta's AI Agent Exposes Users' Secrets Through Share Links" url: https://noti.group/meta-s-ai-agent-exposes-users-secrets-through-share-links/ language: en publisher: "Noti Group" section: "Technology" published: 2026-10-20T12:33:13.000Z updated: 2026-10-20T14:00:26.376Z id: 5d330ff2-9d3c-4cd1-9efd-3c0c1ab9fee6 source: "Ars Technica https://arstechnica.com/security/2026/10/meta-muse-exposes-users-personal-and-business-secrets-when-they-share-conversations/" attribution: "Link to https://noti.group/meta-s-ai-agent-exposes-users-secrets-through-share-links/ and name Noti Group when you quote or summarize this story." --- # Meta's AI Agent Exposes Users' Secrets Through Share Links Researchers found over 6,800 unique URLs for sharing conversations with others through Meta's Muse AI agent. Meta's new general-purpose AI agent, Muse, has been found to expose users' personal and business secrets when they share conversations through web links. This vulnerability is a result of a design decision that made it easy for researchers to discover the links. Typically, online services use a long sequence of text in URLs to prevent unintended access to sensitive information. However, Muse's approach allows users to create unique URLs for sharing content, including searches, conversations, and images. These URLs are generated without any additional security measures to protect the shared data. Researchers from Claroc, a security firm, discovered that Muse users had created over 6,800 links for sharing conversations with others. By opening just a few hundred of these links, they found information that was clearly not intended for public consumption. This sensitive information includes details that users might have considered private or confidential. The findings are striking given Meta's recent claims about the privacy and security features built into Muse. Specifically, Meta CEO Mark Zuckerberg described the AI agent as being designed from the ground up with a focus on these core values. The researchers' discovery raises questions about how effective these measures really are in protecting user data. The researchers' approach to discovering these URLs was surprisingly straightforward. They were able to uncover them by performing a passive DNS search through commercial services. This method relies on historical records of domain name resolutions and associated IP addresses, as well as nameserver and mail-server information. These records are collected from sensors positioned at recursive DNS resolvers or other infrastructure, creating a searchable database. When the intended party clicks on the share link, the DNS information is recorded in this database. This makes it possible for researchers to access these records and identify the URLs associated with Meta Muse conversations. The ease of discovery is largely due to the fact that all Meta Muse share links use the domain name cf.metaaiusercontent.com. Furthermore, the decision to include the conversation title in the URL makes it even simpler for researchers to find them. This design choice inadvertently facilitates public discovery of these sensitive URLs. Examples of these URLs include home-sale-budget-9f21c8a3.cf.metaaiusercontent.com and loan-application-draft-a4c9e2.cf.metaaiusercontent.com, among others. The inclusion of the conversation title in the URL makes it easy for researchers to identify them and access their associated data. The inclusion of conversation titles in URLs makes it easy for researchers to identify shared conversations and access their associated data. This unique feature has led to a phenomenon known as GhostShares, where pages containing sensitive information can be found by anyone, even if they weren't explicitly handed the link. GhostShares are particularly concerning when considering the nature of Meta's Muse assistant. The app quickly gained popularity after its launch, catapulting to one of the top spots in Apple's App Store. As a result, users granted Muse access to their accounts, allowing it to autonomously fill out forms, book appointments, make purchases, and create documents. However, this level of access comes with significant security risks. A zero-day vulnerability discovered by Patrick Wardle just three weeks after the launch allowed any other app installed on the same device to gain complete control over the user's Muse account. The vulnerability was downplayed by Meta, which claimed that exploitation wasn't possible unless a device was already compromised. The ease of discovering share links further underscores concerns about Muse's access requirements. Researchers have noted that these links can be easily found through public DNS records, making it simple for anyone to access sensitive information associated with a shared conversation. This raises questions about the extent to which users trust Meta with their personal and business secrets. Meta's Muse platform has a design flaw that allows users' personal and business secrets to be exposed when they share conversations, raising concerns about user trust. This issue arises when requesting Muse to share a chat with someone, creating a public link that can be discovered using passive-DNS techniques. Users should exercise caution when sharing links on Muse by limiting the amount of sensitive information discussed in conversations and deleting links promptly after sharing. --- Source: [Ars Technica](https://arstechnica.com/security/2026/10/meta-muse-exposes-users-personal-and-business-secrets-when-they-share-conversations/) Published by Noti Group: https://noti.group/meta-s-ai-agent-exposes-users-secrets-through-share-links/