Shared Claude Chats Containing Sensitive Data Briefly Surfaced on Google
An unidentified number of Claude conversations and Artifacts briefly appeared in Google search results, some containing sensitive personal and business information, according to a report by TechCrunch. The issue involved Claude's share chat feature, which creates links accessible to anyone holding the URL.
- Feature involved: Claude's share chat function, which generates a link anyone with the URL can open
- How it surfaced: Reddit users found that search operators such as site:claude.ai/share returned shared conversations
- Reported content: health records, private company documents, and names and phone numbers of children
- Also reported by Futurism: a medical report, clinical trial results containing patient names, documents listing children's names and phone numbers, internal company documents and employee reviews containing personal information
- Timeline: first flagged by a Reddit user on Saturday, reported by 404 Media on Monday morning
- Apparent resolution: by Monday afternoon, a test using the same search method returned no results
What Anthropic Says Happened
Anthropic told TechCrunch that shared links appear in search results only when users post them somewhere search engines can reach, such as forums or social media, and that links shared privately do not appear in search results.
Anthropic spokeswoman Amie Rotherham said the company gives users control over whether they share Claude conversations publicly, and that it does not provide chat directories or sitemaps to search engines. She added that when a user shares a conversation it becomes publicly accessible and can be archived by third-party services.
That account draws a specific distinction: between a system indexing private content, which the company denies, and users publishing links themselves which search engines then crawl. Both descriptions can be true of the same outcome, and the practical exposure for the people named in those conversations is identical either way.
Why This Matters Beyond One Product
Share features of this kind are now standard across AI assistants, and they share a common design assumption: that a link is private because it is unguessable.
That assumption holds only until the link is posted somewhere. Once a URL appears in a public forum, a support thread, a shared document or a social post, it is crawlable, and the content behind it becomes searchable regardless of what the person who created it intended.
For marketing and communications teams the operational point is narrow and immediate. Conversations that include unreleased campaign material, client data, pricing, employee information or anything covered by an NDA should not be shared through a public-link feature, on any assistant, regardless of the platform's indexing policy. The control that matters is what goes into the conversation, not what happens to the link afterwards.
The reported speed of the fix is worth recording alongside the failure. The issue was flagged on a Saturday and appeared to be addressed by Monday afternoon.
More on Ad Tribe Ad Tech.