[David Sacks: Safety Guardrails for Models May Undermine U.S. AI Competitiveness in Cybersecurity] Deep Tide TechFlow reports that on July 20, David Sacks, an AI advisor to the U.S. White House, stated that Kimi K3 had fixed 15 critical security vulnerabilities, while Codex and Fable refused to handle related tasks due to 'safety guardrail' restrictions. He argued that if Chinese models are capable of performing such cybersecurity tasks, but U.S. models continue to face stricter limitations, it could weaken their competitiveness. Sacks also cited an example from Hugging Face, noting that when analyzing an AI-driven cyberattack, a cutting-edge U.S. model was unable to process requests containing real exploit payloads due to safety restrictions. The team ultimately turned to the locally-run GLM 5.2, suggesting that excessive guardrails might actually weaken defensive security capabilities.