Anthropic Cuts Distillation Data Harvesting by 90-99 Percent

Key Findings

Anthropic says its new defenses have cut distillation data harvesting from its model Claude by 90 to 99 percent.

Background

A September 2026 threat intelligence report details how Anthropic tracked, attributed, and disrupted illicit attempts to extract Claude's capabilities. The focus falls mainly on reasoning, coding, and tool use.

Distillation Efforts

Anthropic attributed nearly 200 million distillation-related exchanges across five separate campaigns. The biggest of the five stands out, linked to Alibaba's Qwen, and involved over 151 million exchanges between May and July 2026.

Anthropic's Defenses

Anthropic describes a layered approach to defend against distillation attempts. Account bans sit at one level, followed by metadata analysis and behavioral classifiers. On top, the company uses organizational attribution methods to connect activity back to the groups behind it.

Product Updates

Several of Anthropic's defenses arrived with product updates. The Fable 5 update strengthened extraction classifiers to flag systematic harvesting. It also introduced summarization features in Claude. The Fable 5.1 update added 'preserved thinking,' preventing new API accounts from editing encrypted reasoning context.

Conclusion

Anthropic is not alone in this fight. OpenAI has also been working on disrupting distillation efforts. Illicit distillation raises questions about whether the safeguards built into the original model carry over to the distilled copy. Anthropic draws a line between this kind of activity and legitimate research use.