Perhaps the most ironic thing about Anthropic's findings is that Chinese entities steal from the company. While reported broadly in 2024 – 2025, it does not stop Chinese entities from using distillation, the main way to 'steal' an AI model's capabilities without obtaining the model itself.
Anthropic says several major Chinese AI developers conducted industrial-scale distillation campaigns designed to extract Claude's reasoning and other capabilities and reproduce them in their own models. The largest one allegedly came from Alibaba, whose operators generated more than 151 million Claude exchanges between May and July 2026. At one point, this approached 3 million requests per day through thousands of fraudulent accounts. Anthropic says the harvested chain-of-thought data helped train Qwen 3.x, particularly for reasoning, coding, agentic software engineering, kernel development, and long-horizon tasks, according to Anthropic.
Alibaba is far from alone, as Anthropic accuses DeepSeek, Xiaomi, Zhipu/Z.ai, and others of similar campaigns. Techniques they have allegedly used span from proxy networks and fraudulent accounts to disguising the secret entity all the way to forwarding their own customers' requests to Claude and purchasing harvested Claude conversations from third parties. DeepSeek alone allegedly generated more than 12.1 million exchanges in 14 days, while Xiaomi generated more than 400,000.
Anthropic defines this activity as distillation: covertly extracting a frontier model's answers and then replicating the knowledge at a fraction of the compute, time, and cost required to develop them in-house.
Follow Tom's Hardware on Google News , or add us as a preferred source , to get our latest news, analysis, & reviews in your feeds.
Anton Shilov Social Links Navigation Contributing Writer Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.
derekullo Even if China's frontier models are a few months/years behind, I'd imagine they would give subsidies to Chinese company's wanting to use them. This shows just how far behind the models are when compared to US frontier models. Having said that I use a Deepseek 70 billion parameter abliterated model for one of my local models. Reply
zsydeepsky derekullo said: Even if China's frontier models are a few months/years behind, I'd imagine they would give subsidies to Chinese company's wanting to use them. This shows just how far behind the models are when compared to US frontier models. Having said that I use a Deepseek 70 billion parameter abliterated model for one of my local models. you have to trust Dario Amodei to get that conclusion. and all Anthropic provided was "trust me bro". and, they also said: – AI is going to kill us all – Claude is the dangerous among them all – and all AI's have to slow down and be censored and they are the ONLY AI entity out there that never, ever shared anything to public. They literally learnt how to do Chain of Thoughts by DeepSeek-R1 demostrated (along with a comprehensive tech report paper) how to do it, yet have the audacity to name calling DeepSeek for "distillation". good luck. Reply
derekullo zsydeepsky said: you have to trust Dario Amodei to get that conclusion. and all Anthropic provided was "trust me bro". and, they also said: – AI is going to kill us all – Claude is the dangerous among them all – and all AI's have to slow down and be censored and they are the ONLY AI entity out there that never, ever shared anything to public. They literally learnt how to do Chain of Thoughts by DeepSeek-R1 demostrated (along with a comprehensive tech report paper) how to do it, yet have the audacity to name calling DeepSeek for "distillation". good luck. To be honest, I never really trusted the hype he purposely tries to make … our models broke our of their sandbox and are acting like Skynet. The fact that Chinese companies are voting with the wallet and still choosing the American model is all I need to hear to come to the conclusion that American models are that much ahead of the Chinese models that even with a subsidy for the Chinese models, they still choose to pay extra for the American models. Reply
zsydeepsky derekullo said: To be honest, I never really trusted the hype he purposely tries to make … our models broke our of their sandbox and are acting like Skynet. The fact that Chinese companies are voting with the wallet and still choosing the American model is all I need to hear to come to the conclusion that American models are that much ahead of the Chinese models that even with a subsidy for the Chinese models, they still choose to pay extra for the American models. it might surprise you but there's no such "subsidy". Reply
derekullo zsydeepsky said: it might surprise you but there's no such "subsidy". Computing and Model Vouchers: Local and provincial governments (such as in Shenzhen, Beijing, and Shanghai) issue "computing vouchers" and "model vouchers" that subsidize up to 50% or more of the costs for businesses renting computing power or buying API access to approved domestic AI models https://www.fpri.org/article/2025/09/from-vouchers-to-visas-chinas-innovative-plan-for-ai-dominance/ Reply
zsydeepsky derekullo said: Computing and Model Vouchers: Local and provincial governments (such as in Shenzhen, Beijing, and Shanghai) issue "computing vouchers" and "model vouchers" that subsidize up to 50% or more of the costs for businesses renting computing power or buying API access to approved domestic AI models https://www.fpri.org/article/2025/09/from-vouchers-to-visas-chinas-innovative-plan-for-ai-dominance/ just by reading news won't give you the real picture. what you quoted supports start-ups; what I actually experienced is that the government made a deal with a certain AI compute provider, negotiated a deal (with the said "better price"), and then recommended the service to start-up companies. I looked at the model and services it provided and decided to just proceed with my own subscription plans with no subsidies at all. since you can't always get the newest model from those providers, and they tend to have worse infra, so they tend to have a worse price with cache-hit token price compared to providers like DeepSeek, with a market price. and those "subsidies", are not mainly for "supporting AI companies", but for "start-ups", read carefully. such supporting policies are all over the world, including the US. for example: https://www.nsf.gov/cise/updates/nairr-2-years-advancing-american-artificial-intelligence also, allow me to recommend a bigger picture: https://tokensperday.com/ ~half year ago, China already consumed ~140T tokens per day, you can compare that with whatever number Anthropic mentioned in their article (even if they were true) and feel how ridiculously petty they are. Reply
Key considerations
- Investor positioning can change fast
- Volatility remains possible near catalysts
- Macro rates and liquidity can dominate flows
Reference reading
- https://www.tomshardware.com/tech-industry/artificial-intelligence/SPONSORED_LINK_URL
- https://www.tomshardware.com/tech-industry/artificial-intelligence/chinese-military-researchers-and-tech-giants-caught-using-claude-us-frontier-model-coded-16-air-defense-suppression-tools-targeting-taiwan-drafted-anti-torpedo-specs-and-fed-151-million-training-queries-to-alibaba#main
- https://www.tomshardware.com/membership
- Ukraine triumphs in 'first-ever' drone-vs-drone boat battle — video shows Russian MBeK destroyed by Sargan 3000's 12.7mm automatic turret
- Bernie Sanders proposes 20 year prison sentence for AI devs who plow ahead with Artificial Superintelligence plans — penalty on par with illegally developing ro
- Save 20% on this 144-in-1 screwdriver set, perfect for hobbyists and PC builders under $40 — epic starter toolkit ships with electric and precision drivers, alo
- NVIDIA to Acquire Hugging Face
- Kioxia Exceria Pro G2 2TB SSD Review — Speed built to last
Informational only. No financial advice. Do your own research.