User conversations on the Claude platform were indexed by Google and became visible in search results this Monday, July 27th. The responsible company has already taken measures to remove this indexing, and currently, chats can only be accessed through direct links.
The first reports of the incident emerged on Reddit after a user shared a screenshot showing several conversations listed by the search engine. By using the search command «site:claude.ai/share» in the search bar, the engine returned an extensive list of clickable pages, which led internet users to the complete and authentic transcriptions of interactions with the virtual assistant.
A spokesperson for Anthropic informed the BBC that the conversation links «were not detectable», unless people chose to share them voluntarily. He explained that when a dialogue is shared, the content becomes publicly available and, like other open materials on the web, can be archived by third-party services.
According to the IT Home portal, the exposed material contained sensitive information. Anyone performing the search could read lawyers discussing confidential legal strategies, programmers solving technical issues of corporate systems, and even private conversations with intimate confessions.
Claude has a native feature that allows the creation of a public link for a chat. This feature was designed to aid collaboration, allowing team members to track the progress of a design or software project by receiving the generated URL.
In theory, these pages should be restricted only to the recipients of the link. The address should contain the «noindex» instruction in the source code, which would prevent Google bots from crawling and displaying the page in search results. However, according to reports, some conversations were published without this essential marking. Consequently, any link shared on forums, social networks, or other websites could be normally indexed by Google and included in its public index due to the absence of this single directive in the code.
Although Anthropic has not yet provided details on the scope of the problem or the exact number of affected users, the records have already been removed from the Google index. Currently, searching using the term «site:claude.ai/share» no longer displays the exposed links.
For Claude users, it is recommended to check their own accounts. The process involves accessing settings, navigating to the «Privacy» menu, and entering the «Shared Chats» section. In this panel, you can view all created URLs and deactivate those that no longer require external access.
Hundreds of dialogues generated by users of Claude, the artificial intelligence chatbot developed by Anthropic, became accessible to the public after being found in search engines. This discovery exposed interactions containing personal and professional data, raising important questions about privacy limits when using AI tools.
According to the BBC, the exposure occurred because the users themselves chose to share the links to their Claude conversations. These subsequent addresses were indexed by search engines, such as Google, allowing third parties to find the content. Although the option to locate this material was disabled over the weekend, some of the links had already been saved and distributed on the internet.
An Anthropic spokesperson clarified that the decision to disclose a conversation remains with the user. She detailed that the links in question were not easily detectable or predictable unless there was a voluntary sharing action by the individuals. The company emphasized that by making an interaction public, the individual authorizes external services, analogous to other web content, to access and archive that material.
The initial identification of these dialogues was made by Reddit users, who located over 200 conversations spread across at least 25 pages of search results. The examples of chatbot usage were varied, ranging from creating original texts about cloud security to discussions related to corporate projects.
Furthermore, cases were found of assistance in drafting resumes containing personal data and confidential research related to the workplace, including the healthcare sector. One notable record showed a user asking Claude for help to transform into the Nine-Tailed Fox, and the chatbot responded by generating an AI image and stating that the user had obtained 'fully functional fox powers!'
This incident with Anthropic is not isolated in the scenario of conversation exposure on AI platforms. The BBC mentioned that OpenAI has faced a similar situation with public records of ChatGPT, which led to a modification in how such content is accessed. Grok, the chatbot of the X platform, also had hundreds of thousands of records made publicly available through online searches.
A Google representative informed the BBC that the company does not determine which pages remain public on the internet, as this prerogative belongs to the websites themselves. However, the company provides tools that allow pages to prevent being tracked or indexed. With the removal of Claude's records from search results, it is expected that Anthropic has used the available mechanisms to block these links. The occurrence underscores the need for caution when sharing information, as a conversation generated in an AI tool can quickly leave the private sphere if shared publicly.
