The Chinese AI model Kimi K3 successfully addressed 15 critical security vulnerabilities in a software project, while OpenAI's Codex and Anthropic's Fable declined to fix them due to their internal "cyber guardrails." This incident suggests that self-imposed restrictions on American models may hinder their competitiveness against less constrained Chinese counterparts.
US AI models face a competitive disadvantage in critical tasks if self-imposed guardrails prevent them from addressing real-world problems that Chinese models handle.