Claude's Shared Chats Exposed in Google Search Results

Shared conversation histories from Anthropic's chat service "Claude" were temporarily appearing in Google search results. The cause was the absence of a "noindex tag" setting on the shared pages, which is designed to prevent search engines from indexing web pages. Conversation content that should have remained private was inadvertently accessible to anyone through search.

The "noindex tag" is an HTML setting that instructs search engines not to register a web page in their database (index). Without this tag, search engine crawlers (programs that automatically collect information) will read the page and include it in search results. It is important to note that the issue was not with the sharing feature itself, but rather with inadequate page configuration settings that led to this exposure.

According to user reports, shared chats appearing in search results included conversations containing cryptocurrency-related keys and legal questions. Both types of information were not intended to be disclosed externally. The duration of the exposure and the number of affected chats have not been clarified at this time.

Similar incidents have precedent. OpenAI experienced the same configuration error last year, resulting in shared ChatGPT conversations appearing in search engines. In other words, Anthropic's current issue could be viewed as a mistake that may be repeated across the entire industry.

This incident reveals the magnitude of risk posed by the "sharing feature" in AI chat services. While users can easily share conversations, control over the scope of disclosure is entirely dependent on the service provider's configuration. A single configuration oversight can result in privacy-related information being unintentionally exposed to a broad audience, highlighting a structural problem in how these services operate.

For AI chat users, this case serves as an opportunity to reconsider the nature of information they share. When dealing with information that should not be known to third parties—such as cryptocurrency keys or legal consultations—users should exercise caution when using the sharing feature. It is difficult for users to predict when a service provider's configuration errors might occur, and at the individual level, it is important to adopt the practice of not inputting sensitive information into AI interactions in the first place, especially information that should never be exposed externally.

As AI service usage becomes more routine, addressing privacy risks from such technical configuration errors is positioned as an increasingly important issue for each company. With configuration errors confirmed at both OpenAI and now Anthropic, attention will focus on whether discussions regarding industry-wide security standards and configuration standardization will accelerate in the future.