TechFlow reports that on July 20, White House AI Advisor David Sacks stated that Kimi K3 fixed 15 critical security vulnerabilities, while Codex and Fable refused to handle related tasks due to "cyber guardrails" restrictions. He believes that when Chinese models can complete such cybersecurity tasks, if U.S. models continue to be subject to stricter restrictions, it may weaken their own competitiveness.
Sacks also cited a Hugging Face case, stating that U.S. frontier models, when analyzing an AI-driven cyber attack, could not process requests containing real exploit payloads due to security restrictions, and the team eventually turned to locally run GLM 5.2, believing that excessive guardrails may instead weaken defensive security capabilities.
