AI Pulse · Under the Radar

A small group of archivists is racing to preserve thousands of exposed Claude conversations before Anthropic patches the leak. They’ve already mirrored 1,528 chats containing 16,394 messages, and they’re using TLS-impersonation scrapers to bypass Cloudflare’s protections while Anthropic scrambles to lock things down.

This isn’t the leak story itself. That’s been covered. This is the cottage industry that springs up after a leak, the people who treat exposed data like a historical record worth preserving. The project started with Claude Share links but has now expanded to Grok, Kimi, Qwen, and DeepSeek. The goal is simple: screenshot the leak before it’s fixed. You can see the repository tracking the effort in real time.

Why this matters to you

If your team uses AI assistants, someone is probably sharing links to conversations. Maybe for collaboration, maybe for troubleshooting, maybe just to show a colleague something interesting. Most people assume those links are private or expire quickly. This project proves they don’t always work that way, and when they break, the window stays open long enough for someone to walk through it.

The real lesson isn’t about Anthropic specifically. It’s that share features across AI platforms are built fast and patched later. Your internal processes should assume any shared AI conversation could become public, even temporarily. That means training your team on what not to put in a chat, even one that feels private. It also means having a way to track what’s being shared and where, which is exactly the kind of workflow we build into an AI command centre at Omni.

This is a small story, but it’s a useful reminder. The tools move faster than the guardrails.