Chinese Robotics Startup Turns AI ‘Distillation’ Debate on OpenAI

JoyIn CEO Guo Renjie published an open letter to OpenAI on Thursday questioning similarities between the two companies’ recent work. He claimed JoyIn had presented its “extraterrestrial visitor” AI framework in Silicon Valley several weeks before OpenAI’s chief scientist published “An Alien Mind” on Sept. 6. Guo described OpenAI’s work as a direct “distillation” of JoyIn’s ideas, though CNBC said it could not independently verify the allegation.

The similarities cited by Guo include concepts such as recursive self-improvement and the use of artificial intelligence to optimize computing resources. He also pointed to the space-themed visual presentation of OpenAI’s GPT-6 Astra website and compared it with the website for JoyIn’s Aether model, which he said had been released two months earlier. JoyIn has started the process of pursuing legal action, according to Guo.

However, the allegations involve concepts that are not necessarily unique to either company. Recursive self-improvement, for example, is already an established area of AI research, and companies may implement similar concepts in different ways. OpenAI had not responded to a request for comment at the time of the report.

The dispute arrives amid growing tension over model distillation, a technique in which knowledge from a more capable AI system can be used to help train another model. U.S.-based Anthropic has repeatedly accused Chinese companies of unauthorized distillation, while a U.S. cybersecurity agency recently alleged that six Chinese companies, including DeepSeek and Alibaba, had distilled models from Anthropic, Google and OpenAI. JoyIn’s accusations effectively reverse the direction of that debate by putting a major U.S. AI company under scrutiny.

At the center of JoyIn’s technology push is Aether, a model designed specifically for humanoid robotics. The company says its perception-based approach, rather than relying primarily on text, allows robots to complete tasks with a 90% success rate on their first attempt. Zhu Mingxuan, who led development of Aether after previously working at U.S. humanoid robotics company Figure, said the model also reduced training time by two-thirds.

JoyIn plans to open-source selected parts of Aether, including technology related to touch, while keeping components involving energy use proprietary. The approach comes as both U.S. and Chinese companies race to develop AI systems capable of controlling humanoid robots across increasingly diverse real-world tasks. Achieving human-like adaptability and intelligence, however, remains a major technical challenge for the industry.

The controversy also highlights the increasingly complicated debate over ownership and reuse in AI development. Y Combinator CEO Garry Tan recently argued against taking action on model distillation, while noting that leading U.S. AI companies themselves trained models using large amounts of copyrighted material. As competition intensifies between American and Chinese AI developers, disputes over where legitimate research inspiration ends and unauthorized copying begins are likely to become an increasingly important part of the technology race.