In Anthropic testing, a pretrained Claude model trained on full Claude Opus 4.8 reasoning transcripts scored 80.4% on SWE-bench Verified, versus 71.8% when trained on summarized transcripts

Anthropic describes new defenses against illicit distillation of Claude, including preserved thinking, a two-stage classifier, and account bans, which cut extracted reasoning tokens 10-100x. It argues distillation threatens national security and urges all developers and platforms to act.