Strengthening our defenses against distillation attacks
AI · Security · 2026-10-13 01:46
Anthropic released new defenses that effectively reduce the risk of Claude models being subjected to distillation attacks, with data showing theft reduced by 90% to 99%.