Some Claude conversations and Artifacts shared through public links were found appearing in Google Search results over the weekend, raising concerns about users unintentionally exposing sensitive information.
The issue was first flagged by a Reddit user on Saturday, who found that searches such as “site:claude.ai/share” returned numerous shared Claude conversations.
The total number of affected pages remains unknown. Importantly, there is no evidence that private Claude accounts or conversations were breached. The affected content had previously been turned into publicly accessible links using Claude’s sharing feature.
Some of the publicly searchable conversations reportedly contained highly sensitive information.
Futurism reported finding a detailed medical report belonging to a real patient, clinical trial information containing patient names, internal company documents, employee reviews with personal details, and documents containing the names and phone numbers of primary school-aged children.
Claude Artifacts were also found in search results.
Artifacts are interactive documents, applications, dashboards, code, and other content users can create through Claude. Reports indicated that exposed Artifacts included code, work notes, and business-related documents.
VentureBeat independently found Claude Artifacts belonging to other users through Google Search without already knowing their URLs.
Claude conversations are private by default.
Users can choose to generate a public sharing link for a conversation or Artifact. Claude tells users that anyone with the link can view the content.
The concern is that some users may have understood this to mean that only people who received the difficult-to-guess link would see the conversation, similar to an unlisted link.
Instead, once a link was published somewhere that a search engine could discover, the page could potentially appear in search results and become much easier for strangers to find.
Anthropic said it does not provide directories or sitemaps of shared Claude conversations to Google or other search engines.
According to the company, the randomly generated links are not normally discoverable unless a user posts them somewhere accessible to search engines, such as a public website, forum, or social media platform.
Anthropic spokesperson Amie Rotherham said that when users share a conversation, they make that content publicly accessible, meaning third-party services may archive it like other public web content.
In other words, creating a public link did not automatically make the URL easy to find. But posting that link somewhere publicly accessible could allow search engines to discover and index the conversation.
Google said search engines do not decide which web pages website owners make publicly accessible.
Google spokesperson Ned Adriance said website operators have tools that tell search engines whether pages should be crawled or indexed, and Google respects those instructions.
This distinction is important because the issue was not Google breaking into private Claude conversations. Search engines were indexing pages that were already publicly accessible through their sharing links.
The controversy instead centers on whether Claude users were sufficiently aware that shared links could eventually become discoverable through search engines.
By Monday afternoon, TechCrunch could no longer reproduce the Google search that previously returned shared Claude conversations.
That suggests action was taken to remove or prevent those pages from appearing in Google Search, although it was unclear whether Anthropic, Google or another change was responsible.
Other reports also observed many Claude results disappearing or becoming significantly harder to find by Sunday.
Removing a page from Google Search, however, is different from making the underlying public link private. Anyone who already has an active shared URL may still be able to access it until the owner revokes sharing.
Fortune also reportedly found a conversation labelled “shared by Anthropic” in which Claude produced erotic material.
Anthropic’s usage policies prohibit sexually explicit content.
It remains unclear how the conversation was generated or whether the model was prompted in a way intended to bypass its restrictions. Anthropic had not provided TechCrunch with a specific response about that conversation at the time of the report.
This is not the first time publicly shared AI conversations have appeared in search engines.
Last year, hundreds of Claude conversations were found indexed by search engines. Google estimated at the time that it had indexed just under 600 conversations before the results disappeared.
A separate incident involving ChatGPT also showed the risks of sharing AI conversations publicly. In 2025, a researcher was able to collect around 100,000 publicly shared ChatGPT conversations.
The latest Claude incident again highlights the difference between a conversation being private and a conversation being publicly accessible through an unlisted-looking link.
Claude users can review conversations for which they have created public links by going to:
Settings → Privacy → Shared Chats
Users concerned about previously shared conversations can review that list and disable links they no longer want publicly accessible.
Anyone who shared sensitive information such as passwords, API keys or other credentials through a public conversation should also consider replacing those credentials rather than relying only on removing the shared link.
Get the latest tech news, telecom insights, and product launches wherever you prefer.
Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.