Anthropic Accuses DeepSeek and Other Chinese Companies of Mass Copying Claude Model Technology

American startup Anthropic has published a report accusing Chinese competitors of conducting an unprecedented digital intelligence operation. According to the documentation, DeepSeek, Moonshot AI, and MiniMax created approximately 24,000 fake accounts collectively, through which they generated over 16 million queries to the Claude model. The goal of this operation was so-called distillation, which allows Chinese systems to unlawfully acquire unique features and capabilities of American software. The case takes on geopolitical dimensions as the accusations emerge just one month after similar allegations from OpenAI. Anthropic emphasizes that Chinese companies bypassed regional blocks and usage limits to access the bot's most advanced features. Particularly concerning is the fact that DeepSeek allegedly generated millions of responses on politically sensitive topics, suggesting an attempt to adapt authoritarian models to Western debate standards or detect vulnerabilities in security algorithms. The technological rivalry between the US and China intensified after 2022 when the Joe Biden administration imposed strict export restrictions on advanced semiconductors, aimed at preventing Beijing from developing AI with military potential. Simultaneously with the intellectual property theft accusations, the company DeepSeek has come under scrutiny from the US administration over suspected sanctions evasion. Officials claim the company's latest model was trained using the most powerful Nvidia Blackwell chips, which should not reach China. This suggests the existence of an extensive illegal component trading network that could undermine Washington's efforts in controlling digital technology exports. „These labs targeted Claude's most differentiated capabilities, attempting to create shortcuts for their own product development.” — Anthropic Representative In response to these threats, Anthropic announced the implementation of new user identity verification systems and strengthening protective layers against mass querying of models by automated systems. Experts indicate that without a global agreement on AI training standards, the data theft war will only escalate, pushing innovation aside in favor of reverse engineering and industrial copying of solutions.

Perspektywy mediów: Liberal media emphasize the violation of regulations and unfair market competition that hinders innovation in the technology sector. Conservative media highlight the threat to US national security and the necessity of completely cutting China off from Western technologies.