Chinese Robotics Firm Accuses OpenAI of Copying Its AI Framework

Guo Renjie, CEO of JoyIn, has publicly accused OpenAI of directly copying the company’s recent AI framework for humanoid robots. In a letter written in Chinese and translated by CNBC, Guo stated that OpenAI’s recent release appears to be an unmodified copy of JoyIn’s work, which was presented in Silicon Valley weeks earlier. The accusation marks a significant escalation in the ongoing disputes over AI intellectual property between major U.S. and Chinese tech firms.

Guo specifically pointed to the concept of "recursive self-improvement" and the use of AI to optimize computing power as core technological similarities. He also noted striking parallels in the outer space-inspired design of the websites for both models. JoyIn has stated that it has begun the process of filing a lawsuit against OpenAI, although the U.S. company has not yet responded to requests for comment regarding the allegations.

Accusations of unauthorized model copying

The term "distillation" in AI refers to the practice of using a smaller model to learn from the outputs of a larger, more complex model. This process is often used to speed up training or reduce costs. However, when it involves extracting proprietary logic or code from a competitor’s system without permission, it is considered a form of intellectual property theft. Guo argued that the similarities between JoyIn’s Aether model and OpenAI’s recent releases are too specific to be coincidental, suggesting a direct transfer of technical assets rather than independent development.

This incident is part of a broader pattern of alleged data and model theft. U.S. cybersecurity agencies recently reported that several Chinese companies, including DeepSeek and Alibaba, had distilled models from Anthropic, Google, and OpenAI. While some AI concepts are part of the general research landscape, the specific implementation and combination of features can be proprietary. The lack of clear international standards for AI intellectual property makes these disputes particularly difficult to resolve.

Performance claims and competitive stakes

JoyIn claims its Aether model achieves a 90% success rate on first attempts for robotic tasks, a metric that highlights the practical utility of the system. The model uses a perceptive approach to control rather than relying solely on text-based instructions. Zhu Mingxuan, the lead developer, noted that the new framework reduced training time by two-thirds compared to previous methods. This efficiency is a major commercial advantage, as it lowers the cost and time required to deploy humanoid robots in various industries.

The competitive race between U.S. and Chinese companies to build intelligent humanoids is intensifying. While some industry figures argue that data sharing through distillation is a natural part of technological progress, others view it as a violation of fair competition. The outcome of JoyIn’s legal actions could set a precedent for how AI models are protected and shared globally, potentially affecting the development pace and commercial strategies of startups worldwide.

Uncertainty in legal enforcement

Verifying claims of AI model copying is technically complex and legally ambiguous. CNBC was unable to independently verify the specific similarities Guo cited, and OpenAI has not commented on the matter. The challenge lies in defining what constitutes a unique intellectual property in the realm of machine learning, where many algorithms are derived from open-source research. This ambiguity allows companies to operate in a gray area, where they can adopt similar techniques without necessarily infringing on specific patents or copyrights.

As the debate continues, the focus remains on the practical impact of these technologies. For users and investors, the key takeaway is that the reliability and speed of humanoid robots are improving rapidly, regardless of the legal disputes surrounding their development. However, the lack of clear rules may lead to further conflicts, potentially slowing down innovation or creating fragmented ecosystems of AI models that are incompatible with one another.