Anthropic accuses DeepSeek and other Chinese firms of using Claude to train their AI
Why it matters: This case underscores the urgent need for robust AI governance and security protocols to prevent the exploitation of advanced AI models and safeguard against potential national security threats.
- Anthropic alleges DeepSeek specifically targeted Claude's reasoning abilities and sought to circumvent censorship restrictions, echoing similar accusations from OpenAI.
- DeepSeek, known for its efficient AI models, allegedly engaged in over 150,000 interactions with Claude, while MiniMax and Moonshot had millions more, indicating a large-scale effort.
- Anthropic is urging industry collaboration, regulatory action, and restricted chip access to combat illicit AI distillation, as the practice can bypass existing safety measures and enable misuse by foreign entities (per TechCrunch and NYT Tech).
Anthropic is accusing Chinese AI firms DeepSeek, MiniMax, and Moonshot of illicitly "distilling" its Claude AI model via millions of fraudulent interactions to rapidly advance their own AI capabilities, including generating censorship-safe content. This incident highlights growing concerns over AI model security, intellectual property theft, and the potential for unregulated AI to be used for malicious purposes by authoritarian regimes.
