A screenshot from OpenAI’s video announcing GPT-6 Astra. Credit: OpenAI JoyIn, a Chinese robotics startup, has accused OpenAI of copying its ideas and its website. Its chief executive, Guo Renjie, set out the claims in an open letter in Chinese on the company’s WeChat account, dated 10 September. On the same day, JoyIn announced Aether, a model for controlling humanoid robots. “People often say major tech companies have intelligence networks monitoring the whole internet, this time I believe it, this is a direct distillation of us without any modifications,” Guo wrote, in CNBC’s translation.
Evelyn Cheng reported the letter for CNBC. The outlet said it could not independently verify the claims. OpenAI did not immediately respond to its request for comment. TNW has not verified the claims either.
The three questions JoyIn is based in Suzhou and backed by Ant Group, according to CNBC. The letter puts “three questions” to OpenAI. They cover technical architecture and ideas, the concept of an alien intelligence, and website design. On ideas, Guo says JoyIn presented its “extraterrestrial visitor” framework in Silicon Valley on 16 August.
OpenAI’s chief scientist, Jakub Pachocki, published an essay titled “An Alien Mind” on 6 September. Guo cites similarities on recursive self-improvement and on using AI to optimise computing power, CNBC reported. On design, Guo compares OpenAI’s page for GPT-6 Astra with the Aether website. OpenAI launched Astra on 3 September, according to CNBC.
JoyIn says its own site went live at the end of July, more than two months earlier. The letter names particle effects as the core idea of that design. It describes OpenAI’s page as “pixel-level” copying, in TNW’s translation. What JoyIn says it will do JoyIn’s legal team has “fully launched” legal proceedings and started a lawsuit, the letter says.
CNBC reported that the company has begun the process of filing a lawsuit. The letter also says JoyIn brought forward the release of Aether in response to OpenAI. What Pachocki’s essay says Pachocki’s essay discusses each of the areas Guo names. On recursive self-improvement, he wrote: “Machine recursive self-improvement (RSI) will be at the very core of future scientific discovery.” OpenAI focuses its research on RSI because it believes this is “the only way to remain at the frontier”, he wrote.
On computing, he wrote that “AI will improve the computational substrate itself”. That line links to OpenAI’s results for Jalapeño, its custom chip. The alien idea appears in the title and in the phrase “brought about by an alien intellect exceeding our own”. The essay traces OpenAI’s work on reasoning models to mid-2023.
That year, Pachocki wrote, a research project called “RLSlow” produced the first results. TNW covered the essay when it came out. It also covered the launch of GPT-6 Astra. Pachocki also uses the word distillation.
Explaining why OpenAI hides its models’ chain of thought, he wrote: “A secondary reason for this design was preventing distillation.” Where else the ideas appear CNBC noted that some of the concepts Guo cites are existing parts of AI research. It linked to “When AI builds itself”, an essay on recursive self-improvement from the Anthropic Institute. Marina Favaro and Jack Clark wrote it. It says Anthropic is handing a growing share of AI development to AI systems themselves.
Forbes wrote about the essay on 7 June. What JoyIn claims for Aether JoyIn sums up Aether with three numbers: 200, 0 and 4. According to the letter, the model learned from 200 hours of video of humans and no data from real robots, and it has 4bn parameters. JoyIn claims a success rate above 90% at the first attempt, without task-specific training.
It says the model runs on robots from Unitree and AgiBot. The letter describes an architecture modelled on regions of the brain, from the prefrontal cortex to the spinal cord. It lists five capabilities: active perception, multimodal understanding, streaming memory, motor dynamics and self-iteration. JoyIn says it assembled the 20-person team in March.
The Chinese tech outlet QbitAI reported a ten-minute demonstration filmed in a single take, according to the letter. Zhu Mingxuan led the model’s development. She left the US humanoid maker Figure last year, CNBC reported. Zhu told CNBC that Aether cut training time by two-thirds.
She said JoyIn plans to open-source some parts of the model, such as touch, but not the parts related to energy. TNW has not verified the performance figures. The distillation dispute Distillation means training an AI model on the outputs of another model. US agencies have accused Chinese companies of doing this to American models.
On 8 September, the NSA, CISA and the FBI issued an advisory naming six Chinese firms. They were DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI. The advisory says the firms distilled Claude, GPT, Gemini and Grok models, “likely with Chinese government awareness”. It describes queries amounting to billions of tokens across millions of exchanges.
According to the advisory, some of the traffic ran through proxy services known as “transfer stations”. China’s foreign ministry rejected the accusations. CNBC described JoyIn as flipping the distillation claim on OpenAI.
