The CEO of Ant-backed humanoid robotics startup JoyIn has accused OpenAI of directly distilling the company’s Aether model, citing technical and design similarities between the two companies’ recent systems.
“Major tech companies are often said to have intelligence networks that monitor the entire internet. This time, I believe it. What we are seeing appears to be a direct distillation of our work without modification,” Guo Renjie told CNBC in a public statement Thursday, according to the outlet’s translation from Chinese.
Guo said JoyIn publicly unveiled its “extraterrestrial visitor” AI model framework in Silicon Valley several weeks before OpenAI’s chief scientist published “An Alien Mind” on Sept. 6. He pointed to similarities in approaches such as recursive self-improvement and the use of AI to optimize computing power.
Guo also cited similarities between the space-themed design of OpenAI’s GPT-6 Astra webpage and JoyIn’s Aether model website, which he said had been released two months earlier. He said the company had begun the process of filing a lawsuit.
CNBC could not independently verify Guo’s claims. Many of the concepts involved are already part of broader AI research, although companies may connect and implement them differently. OpenAI did not immediately respond to a request for comment.
Anthropic has repeatedly accused Chinese companies of unauthorized model distillation to strengthen their own AI systems. On Tuesday, a U.S. cybersecurity agency said six Chinese firms, including DeepSeek and Alibaba, had distilled models from Anthropic, Google and OpenAI.
JoyIn says Aether uses a perceptual rather than text-based approach to robotic control, allowing humanoid robots to complete tasks successfully on the first attempt 90% of the time.
Zhu Mingxuan, who led Aether’s development, left humanoid robotics company Figure last year after working on systems that help robots imitate human movements. She told CNBC earlier this week that Aether cut training time by two-thirds. JoyIn plans to open-source parts of the model related to touch, but not components associated with energy use.
Making humanoid robots and AI models perform across diverse tasks at humanlike levels remains difficult as competition between U.S. and Chinese technology companies intensifies.
Y Combinator chief executive Garry Tan said this month that U.S. companies should “do nothing” about model distillation, arguing that they trained their systems on data protected by copyright law.
