Anthropic Claims 90‑99% Reduction in Claude Reasoning Token Theft After New Defense
Current overview (not historical)
Anthropic announced that its latest security measures have cut the theft of Claude’s reasoning tokens by up to 99% in a battle against large‑scale reverse distillation attacks. Attackers had leveraged hundreds of thousands of fake accounts to extract internal inference chains from Claude, prompting Anthropic to ban tens of thousands of suspected accounts. The company also publicly named firms such as Zhipu and Moonshot as involved in extracting Claude’s capabilities. While the exact scope of the breach remains unclear, Anthropic’s claim signals a significant tightening of its defenses against industrial‑scale token theft.
Written with AI from the reporting · updated 10/14/2026
Expand the detailed fact route and sources
Story route
A station-by-station route from the first public evidence onward; every station carries its own evidence and uncertainty.
- 1Unclassified fact stationOct 13, 00:00recorded Oct 14, 22:23single sourcerevised
Anthropic 切断 Claude 推理 token 盗取达 99%
Read original source ↗ The DecoderReplay this moment in the explorerEvidence & revisions · 1 source records
- Subject：
- anthropic
- Action：
- 发布防御措施切断推理 token 盗取
- Object：
- Claude 推理 token
- Occurred (fact side)：
- Oct 13, 00:00
- Published (valid side)：
- Oct 13, 19:06
- First seen (known side)：
- Oct 14, 22:23
Fact stationSingle source · one-sided claimRevisedCopy may have been revised later; time replay is bounded. Layer labels group reading only, not causation between layers.
Uncertainty marks flag where to check the sources; they are not proven causation.
Report timeline
Explore every side through dated source reports.
- The DecoderRelated fact: single sourceSource report revisedAnthropic reduces reasoning token theft from Claude by up to 99% against distillation attacks
Report first recorded Oct 14, 22:23
Anthropic implemented new defenses against large-scale distillation attacks on its Claude models, cutting the volume of extracted reasoning tokens—the model's internal chain-of-thought steps—by 90 to 99 percent. Attackers utilized networks of hundreds of thousands of fake accounts; successful distillation attempts dropped sharply after the defenses went live. Anthropic has restricted access to Claude's internal reasoning steps, deployed a detection classifier, and banned tens of thousands of flagged accounts, naming Zhipu and Moonshot as participants in alleged capability siphoning.
Read original source ↗
Event discussion
Reader opinions are hosted on GitHub Discussions; posting requires a GitHub account. Comments are not verified facts.