OpenAI Accuses DeepSeek of Distillation to Train R1
OpenAI accuses DeepSeek of using distillation techniques to train its R1 model, raising concerns about intellectual property and national security.
In February 2026, OpenAI sent a memo to US lawmakers accusing China's rival DeepSeek of using distillation techniques to train its R1 model, improperly exploiting the capabilities of AI models developed in the United States. The accusation raises important questions about intellectual property and national security.
Distillation and Its Implications
Distillation is an AI practice in which a larger, more complex model (the "teacher model") is used to train a smaller, more efficient model (the "student model"). This process allows the smaller model to achieve performance similar to the original, reducing computational costs and speeding up development. However, when used to replicate competing models without authorization, distillation can violate intellectual property rights and the terms of service set by the developing companies.
OpenAI's Accusations Against DeepSeek
According to OpenAI's memo, DeepSeek allegedly used distillation techniques to train its R1 model, using responses generated by OpenAI's models to improve its own capabilities. This approach was reportedly carried out through sophisticated methods designed to bypass OpenAI's defenses against misuse of its models. OpenAI stressed that such practices pose a commercial threat, since DeepSeek offers its R1 model without charging monthly subscription fees, putting pressure on US companies that have invested heavily in developing advanced AI models and charge for premium services.
National Security Concerns
Beyond the commercial implications, OpenAI raised concerns about national security. DeepSeek's chatbot, in fact, censors results on topics considered controversial by the Chinese government, such as Taiwan and Tiananmen Square. This raises questions about how distillation of AI models could influence the spread of information and potentially enable misuse of the models in high-risk areas such as biology or chemistry.
Reactions and Future Developments
OpenAI's accusations have drawn the attention of other players in the technology sector. Companies such as Google and Anthropic have reported an increase in distillation attacks by Chinese companies, fueling concern about the protection of intellectual property and data security. In addition, the US Navy has banned use of DeepSeek's chatbot among its personnel, highlighting concerns about national security.
In response to the accusations, DeepSeek denied any violation of intellectual property rights and said it developed its R1 model using its own data and methods. The company emphasized the importance of promoting innovation and competition in the AI sector while complying with international regulations.
Conclusions
- Intellectual Property Under Scrutiny: OpenAI's accusations against DeepSeek highlight the challenges of protecting intellectual property rights in the AI sector.
- National Security at Risk: Unauthorized use of AI models raises concerns about the security of sensitive information and the spread of controlled content.
- Need for Regulation: The situation underscores the urgency of developing clear international rules to govern the use and protection of AI models.
- Commercial Implications: Unauthorized distillation practices can affect companies' competitiveness and fairness in the AI market.
- International Cooperation: It is essential to promote cooperation between companies and governments to ensure an AI development environment that is secure and respectful of international law.
This affair highlights the complex dynamics among technological innovation, intellectual property protection, and national security, requiring careful reflection and action from all parties involved.