On September 10, Anthropic disclosed interventions spanning cyber operations, influence campaigns, surveillance, fraud, biological misuse, conventional weapons, and illicit distillation. The company reported that users used Haiku, Sonnet and Opus models across these areas. None of the reported cases involved Fable or Mythos-class models except one illicit-distillation case.
The report presents selected notable cases, not a measure of how common such misuse is. Within the biological category, Anthropic documented five case studies where working scientists used models in ways that could potentially assist weapons development. However, Anthropic emphasized that it does not assert harmful intent by those users, and observed model capability does not constitute proof that real weapons were developed.
Anthropic said it disrupted the reported activity, strengthened safeguards, and shared intelligence with authorities and industry partners where appropriate.