GoKawiilAnthropic reports large-scale distillation attacks by Alibaba, Moonshot AI and DeepSeek on Claude
Anthropic published findings showing five distinct campaigns, mostly linked to Chinese AI labs, that extracted nearly 200 million exchanges from Claude models to train rival systems. The largest, attributed to Alibaba, alone generated 151 million exchanges between May and recent months, using tricks like disguised translation requests to expose the model's hidden reasoning steps.
GoKawiil's interpretation of the reporting above, not reported fact.
Distillation lets smaller, cheaper models inherit the reasoning abilities of costly frontier systems without the R&D investment, undermining the commercial edge that companies like Anthropic depend on. The scale and sophistication described suggest an intensifying arms race between US labs and Chinese competitors over who can legally or illegally capture advanced AI capabilities fastest.
- Anthropic identified five distillation campaigns totaling nearly 200 million suspicious exchanges with Claude.
- The largest campaign, tied to Alibaba, generated 151 million exchanges and is called Anthropic's biggest observed distillation effort.
- Attackers used disguised prompts, like fake translation requests, to expose Claude's hidden chain-of-thought reasoning.
Source: techcrunch.com — Russell Brandom, 2026-09-10
Published there as: “Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.