{"id":94602,"date":"2026-02-24T11:44:39","date_gmt":"2026-02-24T08:44:39","guid":{"rendered":"https:\/\/forklog.com\/en\/?p=94602"},"modified":"2026-02-24T11:46:58","modified_gmt":"2026-02-24T08:46:58","slug":"anthropic-accuses-chinese-ai-labs-of-data-theft","status":"publish","type":"post","link":"https:\/\/forklog.com\/en\/anthropic-accuses-chinese-ai-labs-of-data-theft\/","title":{"rendered":"Anthropic accuses Chinese AI labs of ‘data theft’"},"content":{"rendered":"

Anthropic accused<\/a> three Chinese AI startups \u2014 DeepSeek, Moonshot and MiniMax \u2014 of running a large-scale campaign to use Claude to improve their own models.<\/p>\n

Labs from China generated more than 16 million interactions with the chatbot via roughly 24,000 fraudulent accounts, violating terms of use and regional restrictions.<\/p>\n

\n

\u201cWe have a high degree of confidence linking each campaign to a specific firm based on correlations of IP addresses, request metadata, infrastructure signals and confirmations from partners in the industry. They targeted the most unique capabilities of Claude: agentic reasoning, tool use and programming,\u201d Anthropic said.<\/p>\n<\/blockquote>\n

The firms used distillation \u2014 training a less powerful neural network on the outputs of a stronger one.<\/p>\n

It is a widely used and legitimate method. Leading AI labs regularly distil their own models to create compact, cheaper versions for clients.<\/p>\n

\n

\u201cHowever, it can also be used illegally: competitors improve capabilities at the expense of another\u2019s LLM<\/span> in a fraction of the time and cost compared with developing on their own,\u201d the Anthropic blog says.<\/p>\n<\/blockquote>\n

The company stressed that the window to respond to such \u201ctheft\u201d is narrow, and the threat extends beyond a single firm or region. Addressing it will require swift, coordinated action by industry, regulators and the global AI community.<\/p>\n

Why this is dangerous<\/h3>\n

Anthropic set out the risks. Illegally distilled models do not retain necessary safety mechanisms \u2014 creating national-security concerns.<\/p>\n

American firms are deploying systems to prevent the use of AI in developing biological weapons, conducting malicious cyberattacks and other dangerous acts. Models created through unlawful distillation do not receive such constraints.<\/p>\n

Foreign labs can integrate unprotected capabilities into military and intelligence systems, enabling authoritarian governments to use advanced AI for cyberattacks, disinformation and mass surveillance, the company added.<\/p>\n

Countermeasures<\/h3>\n

Anthropic experts backed export controls<\/a> to preserve US leadership in AI. In their view, distillation attacks undermine these measures, allowing overseas labs to narrow the technology gap.<\/p>\n

\n

\u201cWithout transparency into such attacks, the rapid progress of Chinese labs is mistakenly read as proof that export controls are ineffective. In practice, their achievements largely depend on extracting capabilities from American models, and scaling this approach requires access to cutting-edge chips,\u201d the company blog says.<\/p>\n<\/blockquote>\n

Anthropic listed its own countermeasures:<\/p>\n