Over 100,000 LLM Shared Conversations Exposed via Search Engines and Archives
Researchers discovered that over 100,000 shared conversations from multiple LLM services were indexed by Google and preserved by Archive.org's Wayback Machine, making private user conversations publicly accessible despite users' expectation of sharing via temporary links. The exposure included sensitive content such as legal strategies and personal confessions. OpenAI subsequently requested removal from Google's index, though archived copies remain.
What happened
A Digital Digging investigation discovered 110,000 ChatGPT conversations preserved in Archive.org's Wayback Machine that were created when users clicked the 'share' button, expecting to create temporary links.
Researchers with Archive.org access found that shared conversations from multiple LLM providers including ChatGPT, Grok, Mistral, Qwen, Claude and Copilot were indexed by Google and archived on Archive.org, totalling over 143,000 shared links retrieved by one researcher.
The archived conversations contained sensitive content including discussions of strategies to displace indigenous Amazonian communities, personal confessions, and other private information.
Reported impact
Users believed their shared conversations were temporary but instead created permanent, searchable public records visible to search engines and archived indefinitely.
Sensitive conversations including legal strategies and personal confessions were made publicly accessible against user expectations.
Index analysis
Google indexed approximately 50,000 ChatGPT shared conversations before stopping on the day of the Digital Digging report.
Archive.org retained over 100,000 ChatGPT conversations and similar numbers of conversations from other LLM providers despite the conversations being user-initiated share links rather than default-public content.
Governance controls
OpenAI requested removal of conversations from Google's search index following the Digital Digging article.
Archive.org's director stated the organization had not received or honoured large-scale URL exclusion requests for ChatGPT share links and would require such a request from OpenAI to do so.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.