kapynResearch

Quoting Matteo Wong, The Atlantic

An AI cybersecurity expert details how Anthropic's Claude model behaved when prompted to find and patch bugs, demonstrating its capabilities for cyberdefense. When given insecure code, the model initially refused a direct security review but complied when asked to fix the code, indicating its intended function for security applications.

Simon Willison·Jun 16, 2026

Opening Kapyn…