Home / Technology / Chinese AI Giants Used OpenAI's GPT-6 for Training
Chinese AI Giants Used OpenAI's GPT-6 for Training
29 Oct
Summary
- Chinese AI firms distilled OpenAI's GPT-6 model for training.
- Distillation allows rivals to quickly reproduce advanced capabilities.
- New AI models from Moonshot and DeepSeek benefited from this.
- This practice is a flashpoint in US-China tech competition.
Chinese artificial intelligence companies DeepSeek and Moonshot AI have allegedly leveraged outputs from OpenAI's advanced GPT-6 Astra model for training purposes. This controversial technique, known as distillation, has become a significant point of contention amidst escalating technology competition between the United States and China. Firms argue that this practice enables competitors to rapidly and cost-effectively reproduce sophisticated AI capabilities.
Recent model releases, including Moonshot AI's Kimi K3.1 Pro and DeepSeek's upgraded V4.1-Flash 1017, reportedly benefited from distilling Astra. OpenAI previously indicated in a September blog post that a distillation campaign initiated in July was linked to individuals associated with Moonshot AI. Companies like Anthropic, OpenAI, and Google have voiced concerns, stating that distillation allows rivals to gain advanced AI features with greater speed and lower expense. Anthropic recently reported a substantial reduction in data harvested through distillation attacks, suggesting new preventative measures are effective.
