A Chinese robotics startup has flipped a familiar accusation back onto OpenAI. Suzhou-based JoyIn accused the US giant of copying its AI framework. The claim inverts the usual direction of distillation accusations entirely. Typically, Western labs accuse Chinese firms of such copying.
JoyIn CEO Guo Renjie made the claim in a public letter. He published it in Chinese, according to a CNBC translation. Guo pointed to similarities between the two companies’ recent models. He called it “a direct distillation of us without any modifications.”
Distillation refers to extracting one model’s capabilities to train another. Guo’s central argument rested on a specific timeline. He said JoyIn presented its “extraterrestrial visitor” framework recently. That presentation happened in Silicon Valley a few weeks ago.
Crucially, that came before an OpenAI publication he flagged. OpenAI’s Chief Scientist published “An Alien Mind” on September 6. Guo questioned similarities in core technological approaches between them. These included “recursive self-improvement” and AI optimizing computing power.
He also highlighted striking similarities in visual web design. Specifically, he compared OpenAI’s GPT-6 Astra webpage design. He claimed it resembled the website for JoyIn’s Aether model. Both leaned heavily on an outer-space-inspired visual aesthetic.
According to the official blogpost:
While reviewing industry news, we noticed a blemish on OpenAI’s recent flurry of activity—the launch of the GPT-6 Astra website on September 3rd.
On September 6, Chief Scientist Jakub Pachocki published a lengthy article of over 10,000 words.
The article “An Alien Mind” (https://openai.com/index/an-alien-mind/) is presented with great conviction, but many of its elements bear a striking, even absurd, resemblance to our team’s publicly released findings.
My team is confused. Out of respect for the intellectual contributions of team members, and out of a commitment to fair rules in the AI industry.
The accusation carries obvious irony given recent industry events. US-based Anthropic repeatedly flagged unauthorized distillation by Chinese firms. Just this week, a US agency accused six Chinese companies. These included DeepSeek and Alibaba distilling Western models illicitly.
JoyIn timed its own model announcement alongside the accusations. The startup formally unveiled its Aether robotics model on Thursday. Notably, Aether uses a perceptive rather than text-based approach. This method controls humanoid robots through direct perception instead.
The company makes bold performance claims for the new model. It says Aether completes tasks with 90% first-attempt success. This targets a key challenge in humanoid robotics reliability. Such consistency remains difficult for many competing robots today.
The model’s development leader brings notable Western experience too. Zhu Mingxuan led the Aether model’s core development work. She reportedly left US humanoid company Figure last year. There, she also worked on robots mimicking human actions.
In an ironic twist, Anthropic has laid claims that not just one, but three Chinese companies are ripping Claude off, making it a strange twist of blatant copying features in the AI world as of writing this.
