AI Pulse · Frontier Labs Watch

Anthropic’s Claude just had another privacy stumble with its public share-link feature. Users who shared conversation links discovered that anyone with the URL could see their prompts and Claude’s responses, even if those chats contained sensitive business information or proprietary data. The company has patched the issue, but this marks the second time the same problem has surfaced.

For a company that built its reputation on safety and responsible AI, repeated privacy failures are more than embarrassing. They erode trust, especially now that Claude Opus 5 is positioning itself as an enterprise tool meant to handle workflows that touch customer data, strategic planning, and internal operations. If your team is drafting contracts, analysing financials, or brainstorming product ideas in Claude, a share-link leak isn’t a hypothetical risk. It’s a real exposure.

The timing matters. Enterprises are starting to bake AI assistants into daily operations, not just use them for one-off tasks. That means the stakes for access control, audit trails, and data governance are higher. A tool that leaks shared conversations twice suggests process gaps in how features get tested and rolled out. It also raises a harder question: what other edge cases haven’t been caught yet?

If you’re running Claude in your business, review how your team uses share links and what data flows through them. Better yet, think about whether you have visibility into where AI conversations go and who can access them. This is exactly the kind of control layer we build into an AI command centre, so you’re not relying on a vendor’s patch cycle to know your data is contained. The broader lesson is simple: safety marketing doesn’t replace operational rigour, and repeated mistakes cost more than credibility. They cost deals.