OpenAI Says It Disrupted Organized AI Distillation Campaign, Points to Kimi Developer

On September 30, OpenAI posted on its official website that it had recently encountered an organized adversarial distillation campaign. OpenAI said the activity began in the first week of July and matched the characteristics of adversarial distillation, in which another model is trained, reproduced, or improved through systematic unauthorized use of a model's outputs or reasoning processes.

OpenAI revealed that the initial activity on July 1 was small in scale, but traffic surged sharply on July 24 and July 25, when more than 4,000 users made 16,000 requests. Further investigation found that over 15,000 users engaged in similar prompt activity. By July 28, OpenAI had fully blocked the campaign.

OpenAI stressed that the incident did not involve breaking its encryption, breaching databases, or directly accessing user conversations, but the actions violated its terms of service. The company has shared information with industry partners through the Frontier Model Forum to strengthen defenses against distillation.

OpenAI added that it currently does not know the attacking entity, but believes the core activity group is linked to Moonshot AI, the developer of the Kimi model.