Chinese Robotics Startup JoyIn Accuses OpenAI of Copying AI Model Ideas and Website Design

Chinese Humanoid Robotics Startup Challenges OpenAI Over Alleged Copying

JoyIn, an Ant Group-backed Chinese humanoid robotics startup based in Suzhou, has accused OpenAI of copying its technical concepts and website design in a public letter published by CEO Guo Renjie on September 10.1

1

The accusations center on similarities between JoyIn's Aether model and OpenAI's GPT-6 Astra, marking the first known case of a Chinese developer accusing an American company of model distillation.2

2

Guo Renjie claimed that OpenAI's approach represents "a direct distillation of us without any modifications," pointing to similarities in core technological approaches such as recursive self-improvement and the use of AI to optimize computing power.1

1

The CEO stated that JoyIn publicly presented its "extraterrestrial visitor" AI model framework in Silicon Valley on August 16, weeks before OpenAI Chief Scientist Jakub Pachocki published "An Alien Mind" on September 6.3

3

Technical and Design Similarities Spark Intellectual Property Violations Concerns

The accusations extend beyond technical concepts to visual design elements. Guo highlighted what he described as "pixel-level" copying between OpenAI's GPT-6 Astra webpage, launched on September 3, and JoyIn's Aether model website, which the company claims went live at the end of July.3

3

Both sites feature outer space-inspired designs with particle effects as a core visual element.

JoyIn's legal team has initiated the process of filing a lawsuit, according to the letter.1

1

CNBC was unable to independently verify the claims, and OpenAI did not immediately respond to requests for comment. Some concepts cited by Guo, such as recursive self-improvement, are existing parts of AI research more broadly, though companies connect and implement them differently.1

1

Aether Model Claims 90% Success Rate in Humanoid Robot Control

Guo announced that he decided to publicly release the Aether model on September 10 in response to the alleged copying by OpenAI.1

1

The Aether model uses a perceptive, rather than text-based, approach to humanoid robot control, claiming to enable humanoid robots to complete tasks with a 90% success rate on first attempt.1

1

Zhu Mingxuan, who led the Aether model's development, left U.S. humanoid company Figure last year, where she had worked on models for helping humanoid robots mimic human actions.1

1

She told CNBC that Aether reduced training time by two-thirds and that JoyIn plans to open-source parts of the model relating to touch, but not portions related to energy use.1

1

The model reportedly learned from 200 hours of video of humans and no data from real robots, and has 4 billion parameters.3

3

JoyIn assembled the 20-person team in March and demonstrated the model's capabilities in a ten-minute single-take video.3

3

U.S.-China Tech Race Intensifies Amid Model Distillation Disputes

The accusations arrive amid heightened tensions in the U.S.-China tech race over AI innovation and copyright infringement. On September 8, the U.S. Cybersecurity & Infrastructure Security Agency (CISA) stated that six Chinese companies, including DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.ai, are conducting "systematic extraction of proprietary functionalities and capabilities of U.S. AI companies' models through industrial-scale knowledge distillation campaigns."2

2

CISA claimed these efforts form the core of their AI development strategy and are likely carried out with the knowledge and blessing of the Chinese government.2

2

Anthropic also reported on September 11 that Moonshot, developer of the Kimi AI model, had been using a mechanical Turk-style approach to make customers think they were receiving responses from Kimi when they were actually generated by Claude.2

2

Fair Use Debates and Industry Hypocrisy in AI Development

The irony of AI companies calling foul on model distillation when their business models depend on nonconsensual scraping of online content has not escaped industry observers.2

2

OpenAI has faced multiple lawsuits since ChatGPT's launch, filed by artists and news publishers arguing their copyrighted work was used illegally to train proprietary AI models. The company has defended itself by invoking the fair use standard of U.S. copyright law.2

2

Garry Tan, chief executive of Y Combinator, told CNBC this month that he would "do nothing" about distillation, joining others in pointing out that U.S. companies have trained their AI models on data covered under copyright law.1

1

The line between fair use and theft within the unregulated AI industry remains as hazy as the inner workings of the models themselves.2

2

This dispute signals a potential shift in the distillation narrative, where accusations previously flowed one direction from U.S. to Chinese companies. Watch for how legal proceedings unfold and whether this case establishes new precedents for intellectual property protection in AI development across borders.