Anthropic accused three Chinese AI companies of trying to use Claude to improve their models
Anthropic said that three Chinese companies created more than 24,000 fake accounts for its AI chatbot Claude to improve their own models.
Anthropic said that three Chinese companies created more than 24,000 fake accounts for its AI chatbot Claude to improve their own models.
DeepSeek, Moonshot AI, and MiniMax Labs allegedly generated over 16 million chats with Claude through these accounts, using a technique called «distillation,» TechCrunch reports .
Anthropic said the labs «targeted Claude’s most differentiated capabilities: agentic reasoning, tool use, and coding.»
Distillation is a common training method that AI labs use for their models to create smaller, cheaper versions, but competitors can use it to essentially copy other labs’ work. OpenAI sent a memo to House lawmakers earlier this month accusing DeepSeek of using distillation to mimic its products.
The scale of each attack varied in scope. Anthropic tracked over 150,000 chats from DeepSeek that appeared to be aimed at improving underlying logic and consensus, particularly around censorship-proof alternatives for politically sensitive queries.
Moonshot AI had over 3.4 million chats focused on agent-based reasoning and tooling, coding and data analysis, agent development for computer use, and computer vision.
13 million MiniMax chats were targeted at agent coding, tooling, and orchestration. Anthropic said it was able to observe MiniMax performance as it redirected nearly half of its traffic to extract capabilities from the latest Claude model after its launch.
Anthropic says it will continue to invest in defenses that make distillation attacks harder to execute and easier to detect, but calls for a «coordinated response from the AI industry, cloud providers, and policymakers.» The company noted that the scale of mining that DeepSeek, MiniMax, and Moonshot have been performing «requires access to advanced chips.»
«Distillation attacks therefore reinforce the rationale for export controls: restricted access to chips limits both direct model training and the scale of illicit distillation,» the Anthropic blog says.
Anthropic also noted that distillation not only threatens to undermine US dominance in AI, but could also create national security risks.
Read the country’s main IT news in our Telegram
Read the country’s main IT news in our Telegram
«Those who don’t have children lack interest in the future.» Musk criticized the philosopher hired by Anthropic
«Those who don’t have children lack interest in the future.» Musk criticized the philosopher hired by Anthropic
ByteDance’s AI video model — Seedance 2.0 — goes viral: Chinese already predict its success DeepSeek
ByteDance’s AI video model — Seedance 2.0 — goes viral: Chinese already predict its success DeepSeek
Anthropic’s market value rises to $380 billion thanks to new $30 billion investment
Anthropic’s market value rises to $380 billion thanks to new $30 billion investment
