• by Laurie Sullivan , Staff Writer, 10 hours ago

Claude AI leaked everything from personal and financial information to highly sensitive corporate documents, medical reports, cryptocurrency wallet keys, and bizarre personal prompts into Google Search results before the company plugged the hole. 

A Reddit user reported that typing a search operator code used to find publicly indexed links generated by Claude AI's chat feature revealed hundreds of Claude chats, including some surprising conversations that include first names or handles of each affected user.

The ClaudeAI conversations were publicly accessible and allowed anyone to view what seemed like private chats and sensitive information. Some users said they were also able to find the information on Bing and Brave.

This issue, which also affected Claude's "Artifacts" feature, has since been patched, with links no longer appearing in Google search results.

Artifacts are interactive work spaces for trying out vibe-coded apps or other tools. 

"While a few people argued it was "user error" for sharing links, the overwhelming sentiment was that Anthropic should have used a noindex tag to prevent this, especially since OpenAI made the same mistake last year." one Reddit user wrote. "'Share with a link' isn't supposed to mean 'broadcast on Google.'"

Driving the news: During the weekend, users on Reddit and X began warning that they could use Google Search to find Claude conversations, along with apps, documents, spreadsheets, web pages and other files created with Claude.

As of Monday afternoon, many were still able to find publicly shared files indexed on Google, but not Claude conversations people had with the chatbot.

The data exposure is similar to one that impacted OpenAI’s ChatGPT last year.

One Redditor wrote about finding information about a crypto wallet being created that showed the keys and other personal information associated with it.

The conversation eventually reminisced about years ago having the ability to pick up random landline or cell-phone calls on scanners.

"You used to be able to pick up random cell phone chatter on walkie talkies sometimes," wrote one person.

Another wrote it felt like the beginning of a horror movie or some cool indie sci-fi film where you start hearing conversations about the future or an alternate reality.

This is not the first time we have seen Anthropic's Claude leak data related to chats into Google Search. At the time, similar snafus occurred in OpenAI and xAI.

The first time in September 2025, the leaks were made possible through sharing features, but this doesn't appear to be the case this time.

Last year, nearly 100,00 ChatGPT conversations were searchable on Google. 

Google began offering tools that site owners can use to indicate they did not wish certain pages and URLs to appear in search results. But here we are again, with all types of information spilling into Google search results.