Public LLM Share Links Found on archive.org

Date: August 8, 2025

Source event - “I scraped every public LLM chat, so you didn’t have to” (Aug 3, 2025) by dead1nfluence

Source URL - https://julian-b.xyz/posts/llmChatReview.html?

Summary

Recent analysis shows that public shareable chat links from multiple large language model (LLM) providers were captured by Archive.org (Wayback Machine) and can be mass-collected via the Timemap API.

A weekend collection effort by Julian retrieved ~143,142 share links across six major providers (ChatGPT, Grok, Mistral, Qwen, Claude, Copilot). Many of these links resolve directly to chat content or JSON APIs; several are shielded by Cloudflare but remain retrievable using commodity tooling.

Preliminary review discovered confidential information (e.g., AWS Access Key IDs, Replicate tokens) and personally identifiable information (PII), including full names in some Claude shares.

This represents a material risk for organizations whose staff used share links of internal content. Since snapshots persist in the public archives, removal is not guaranteed

It’s best to NOT share LLM share links.

Scope of findings

• Focused on six providers: ChatGPT, Grok, Mistral, Qwen, Claude, Copilot.

• Collection used Archive.org Timemap API for prefix-matching on provider share paths.

• Content retrieval methods varied by provider; some required API endpoint discovery and/or Cloudflare challenge handling.

Key Findings

• Large-scale discoverability: Chat share links can be found via Archive.org API, yielding six-figure volumes of chats.

• Heterogeneous retrieval paths: Some shares return HTML; others call JSON APIs behind the page. Endpoint patterns are consistent enough to script.

• Cloudflare mitigation is insufficient: Tools like Cloudscraper bypass basic bot checks.

• Secrets & PII exposure: Regex scans surfaced cloud/API credentials and identifiable user info.

• Indexing persists: Even if search engines or providers delist/remove links, archived copies often remain accessible.

 To learn more please visit Julian’s link above.