Story perspectives

Anthropic Cuts Claude Distillation Data by 90–99%

By Drooid · Oct 13, 2026, 10:41 AM UTC · How we work

Summary

Story summary

Anthropic announced defenses that reduce Claude model distillation data by 90% to 99%.
The safeguards lower extracted reasoning tokens to one-tenth to one-hundredth of previous levels.
Anthropic froze tens of thousands of distillation-linked accounts, citing Chinese startup Zhipu AI.
Claude Fable 5.1 and Opus 5.5 add a “preserved thinking” feature that blocks context editing for accounts created after Aug 31 2026.
OpenAI reported blocking a distillation campaign on Sep 30, highlighting industry-wide U.S.–China AI competition concerns.

History

Historical Context

OpenAI, founded in 2015 by Sam Altman and Elon Musk, first withheld most of GPT-2 in February 2019 over misuse fears. That caution foreshadowed later U.S.–China AI tensions, as the DoD awarded OpenAI a $200 million contract in June 2025 and labeled rival Anthropic a supply-chain risk in February 2026. The 2019 delay thus became a reference point for emerging bilateral safeguards.

1 of 2

Story summary

Anthropic announced defenses that reduce Claude model distillation data by 90% to 99%.
The safeguards lower extracted reasoning tokens to one-tenth to one-hundredth of previous levels.
Anthropic froze tens of thousands of distillation-linked accounts, citing Chinese startup Zhipu AI.
Claude Fable 5.1 and Opus 5.5 add a “preserved thinking” feature that blocks context editing for accounts created after Aug 31 2026.
OpenAI reported blocking a distillation campaign on Sep 30, highlighting industry-wide U.S.–China AI competition concerns.

1 / 2