How to secure Claude chats after sharing incident

Users of Anthropic’s AI chatbot, Claude, may have concerns about where their data ends up. Over the weekend, they discovered that private conversations were accessible via Google search results. The issue stemmed from a specific search operator that allowed anyone to find public links to Claude chats. By searching for a keyword followed by “site:claude.ai/share,” a vast archive of user interactions became visible. It was a significant oversight. This error exposed the inner workings of how shared links are handled on the platform.

Anthropic has since moved to close this security gap. A search for the specific URL string now yields no results, suggesting the company has successfully blocked Google’s crawlers from indexing these pages. While this is positive news, it is not entirely comforting to know that Google indexed these conversations in the first place. The incident marks the second time in a year that such an event has occurred. A similar situation took place last September. While the developer is not alone in facing this issue—Google once indexed ChatGPT conversations as well—the recurrence raises questions about long-term data privacy.

Why Your Conversations Appeared Online

The exposure does not appear to be a result of a direct data breach where Anthropic handed over user data. Instead, the vast majority of indexed chats were likely conversations that users chose to share publicly. Claude allows people to generate a public link for a chat, a feature designed for sharing specific discussions with colleagues or friends. When a chat is shared, it creates a static page hosted on the web. If the search giant’s bots crawl the open web, they can find and index these pages. This makes them visible to anyone searching for the right terms.

There is a gap between user expectation and technical reality. Many people treat a “share” button like sending a private message, failing to realize that a public link is effectively a published webpage. Until platforms make this distinction clearer with explicit warnings or visual cues, users will continue to unknowingly broadcast their private interactions to the open web. They assume these messages are limited to a specific recipient.

Private chats that were never explicitly shared are unlikely to be involved in this exposure. If you never used the share function, your history should remain secure. However, there is at least one reported instance from the September event where a person claimed their chat was exposed despite never sharing it. The mechanics of that specific anomaly remain unclear.

Certain types of content generated by Claude may still be visible, adding to the privacy concerns. The platform creates “artifacts” when the content meets specific criteria. These can include documents, code snippets, HTML sites, SVG images, diagrams, flowcharts, or Interactive React components. While users often intend for these artifacts to be public, those wishing to keep them private may find they are still indexed on Google even as the main chat links are removed.

Steps to Remove Chats From the Web

While there doesn’t appear to be any reliable way to identify whether your chats and artifacts were included in this exposure, you can ensure that none of them are ever indexed by a search engine in the future. To do so, you’ll essentially have to shut down sharing. To start, head to Settings > Privacy in Claude, then choose “Manage” next to “Shared chats.” Once this new page appears, you’ll see all shared chats, including a link to the conversation and the date shared. From here, scroll through the list and select “Unshare” next to each chat you’d like to remove. This won’t delete your chats; only their snapshots connected to these public links.