OpenAI says Moonshot tried to copy its AI models
What's the story
OpenAI has accused Moonshot AI, a Chinese artificial intelligence (AI) lab, of running a large-scale distillation campaign to copy its AI models. The company flagged the alleged operation in a blog post, saying it tried to extract the protected reasoning of OpenAI's systems. Distilling is when a smaller model is trained on the outputs or reasoning of a larger one.
Accusation details
Moonshot AI has not yet responded to the allegations
OpenAI claims that a "core cluster" of the distillation campaign was linked to Moonshot AI. This suggests that the Chinese lab may have used OpenAI's models for training its future ones.
However, it's worth noting that Moonshot AI has not yet responded to these allegations.
The last update from the lab was on July 16, 2026, when it released Kimi K3.
Counteraction measures
Campaign likely began on July 1, 2026
OpenAI said the operators of this campaign didn't break their encryption or gain direct access to stored user conversations.
However, they did try copying encrypted reasoning from one conversation and getting another model in a different conversation to decrypt and transcribe that hidden reasoning content.
The company said the campaign likely began on July 1, 2026, with more than 4,000 users making about 16,000 requests using an extraction pattern on July 24-25.
User participation
OpenAI was able to completely disrupt campaign by July 28
OpenAI discovered over 15,000 users making similar requests. The company was able to completely disrupt the campaign by July 28.
The activity in this campaign changed over time, suggesting that distillation requires layered and adaptive defenses.
In response to this operation, OpenAI banned or restricted fraudulent accounts and strengthened signup and infrastructure controls while expanding monitoring for related networks.
Safety implications
Distillation campaigns could give AI labs advanced capabilities
OpenAI has warned that such distillation campaigns could give an AI lab advanced capabilities in its models without the same investment in safety.
This is especially concerning as AI safety continues to be a global debate.
Notably, this isn't the first time a US AI lab has accused Chinese AI companies of trying to distill its models.
Anthropic also made similar allegations against Moonshot and others earlier this year.