0
Applied AI·October 13, 2026·1 min read

Anthropic says new measures like preserved thinking in Fable 5.1 and Opus 5.5 have reduced data harvesting through distillation attacks by about 90% to 99%

Share

If Anthropic’s reported 90–99% reduction in distillation attacks holds up, model IP protection just became a lot more technically tractable. Enterprises fine-tuning or exposing APIs should be asking every vendor what their anti-distillation posture is and updating threat models accordingly.