LatticeFlow AI, which measures the regulatory compliance of AI models, says DeepSeek's R1 ranks the lowest in cybersecurity among leading AI models (Parmy Olson/Bloomberg)