Everything on Security & Geopolitics

Large-scale distillation attacks by Chinese labs have been detected, where fraudulent accounts were used to extract data from Claude models. This demonstrates that software-based knowledge transfer can bypass hardware export controls.

Impact: Undermines the effectiveness of chip export restrictions and forces Western AI labs to implement stricter API security and monitoring to protect proprietary model capabilities.

— from AI Infrastructure Wars and Model Efficiency Shifts · Last Week in AI· Mar 03, 2026