Cloudflare's AI-Generated OAuth Library: A Double-Edged Sword

Cloudflare built a new OAuth provider library almost entirely using Anthropic's Claude LLM. While the code is well-structured and tests pass, the author found security issues, such as overly permissive CORS settings, missing standard security headers, and incorrect OAuth spec implementation. Despite engineer review of Claude's output, critical vulnerabilities remain, highlighting the risks of AI-generated code even under scrutiny. This raises questions about the reliability and security of AI-assisted programming and the crucial role of human expertise in security-critical systems.