Anthropic has accused several China-based companies of using its AI model Claude without authorization. This immediately re-ignited debates over AI ethics, intellectual property, and competitive control.
Moreover, the allegations center on so-called "distillation attacks," a practice that can replicate AI capabilities through illicit means.
Understanding Distillation Attacks in AI
Gizmodo explains that distillation is a standard AI process where a "teacher" model provides outputs that a "student" model uses to learn, often producing smaller or more efficient AI systems.
Anthropic distinguishes distillation attacks as attempts to extract model knowledge without permission, bypassing legal and contractual safeguards.
According to Anthropic's Monday blog post, Shanghai-based companies MiniMax, Moonshot, and DeepSeek conducted such attacks. MiniMax reportedly processed more than 13 million exchanges, while Moonshot and DeepSeek processed 3.4 million and 150,000, respectively.
Undoubtedly, these activities allegedly violated service terms and regional access rules. They also sparked concerns over ethical AI deployment and intellectual property protection.
Legal and Ethical Implications of 'Distillation'
Anthropic emphasized that these actions are not criminal but constitute breaches of contractual agreements and U.S. export controls. Circumventing restrictions allows foreign labs, including those linked to government influence, to erode competitive advantages intentionally designed to safeguard American AI innovations.
OpenAI has also raised alarms over similar practices, accusing DeepSeek of "free-riding" on U.S.-based AI research.
AI Volatility
DeepSeek is set to launch its new flagship model, DeepSeek V4, imminently. Analysts warn that its release could increase volatility in AI-driven markets, especially on Wall Street, where investor sensitivity to emerging technologies remains high.
With China's AI sector expected to boom, the AI industry in the country would remain high for the next few years.
Of course, Anthropic should also be consistent in its claims against Chinese AI firms while dealing with Claude's ethical limits on military use.
Originally published on Tech Times
