Hundreds of private conversations between users and Anthropic’s Claude AI chatbot were discovered publicly accessible through major search engines, exposing sensitive personal and corporate information. The breach, which affected more than 200 chat logs across at least 25 search result pages, has reignited debate about privacy safeguards in consumer AI products.
The exposed conversations included users sharing names, contact details, CVs, and in some cases proprietary corporate research and healthcare data. One chat log from April showed a user asking Claude to draft an unpublished blog post about cloud security for a corporate project, including sensitive internal details. Another involved a user seeking help with a job application, listing their full employment history and personal contact information.
Anthropic said the issue stemmed from Claude’s share feature, which allows users to generate a link to any conversation. In its privacy notice, the company warns that anyone with the link can view the contents. However, critics argue the company failed to adequately warn users that these links could be indexed by search engines and become accessible to anyone online.
The search availability was removed over the weekend, but not before the chat logs were saved and circulated widely across social media platforms and forums. The incident mirrors an almost identical problem OpenAI faced last year with ChatGPT, when users discovered that private chat histories were also being indexed by Google and other search engines.
Google told the BBC it does not control what pages websites make public, and that it provides site owners with clear tools to prevent pages from being crawled or indexed. A Google spokesman said the company respects those directives when they are set by site owners, but cannot act preemptively on content it does not know about.
Privacy experts said the incident highlights a broader pattern across the AI industry. Users often assume their interactions with chatbots are fully private, but the architecture of share features, link generation, and search engine indexing creates risks that are not clearly communicated. Unlike email or messaging platforms, where users expect end-to-end encryption and non-indexable content, AI chat platforms operate on a different privacy model that is poorly understood by consumers.
The exposed conversations also included more whimsical exchanges, such as a user asking Claude how to literally transform into a nine-tailed fox, to which the chatbot initially responded by generating an AI image claiming the user had been given fox powers. But the vast majority of exposed chats contained real personal and professional data, including transcripts of private meetings and confidential work documents.
Anthropic said Claude users maintain control over if and when to share conversations. However, the company did not directly address why shared links were not configured with a noindex tag to prevent search engine crawling, a standard practice for private or user-specific web content. The company declined to say whether it would change its default sharing settings in response to the incident.
Lawmakers in Europe and the United States have been increasingly focused on AI transparency and data protection. The Claude exposure is likely to fuel calls for mandatory privacy impact assessments before AI products are released to the public, particularly those that handle user-generated content with share functionality.
The incident also raises practical questions for businesses that use Claude and similar AI tools for internal operations. Several of the exposed chats appeared to involve employees conducting proprietary research or analyzing confidential data through the chatbot, unaware that their conversations could become searchable public records if shared internally or with colleagues.
Cybersecurity experts recommend that companies implement strict policies around the use of consumer AI chatbots for work purposes, including bans on sharing internal data through such platforms unless enterprise-grade privacy controls are in place. The Claude leak serves as a reminder that convenience and privacy often pull in opposite directions in the rapidly evolving AI landscape.