Report Claims Chinese AI Firms DeepSeek and Moonshot Used OpenAI Model for Training

Chinese artificial intelligence developers DeepSeek and Moonshot AI reportedly utilized outputs from OpenAI's GPT-6 Astra model to train their recent models. The practice, known as distillation, has intensified discussions around intellectual property and technology competition between US and Chinese firms.

Chinese artificial intelligence developers DeepSeek and Moonshot AI relied heavily on outputs from OpenAI's GPT-6 Astra model to train their latest models, according to a report published on Wednesday by The Information, citing people with direct knowledge of the matter.

The practice, commonly referred to as distillation, has emerged as a central flashpoint for leading artificial intelligence companies amid intensifying technology competition between the United States and China. Industry participants note that distillation allows rival firms to reproduce advanced technological capabilities more quickly and at a lower cost than developing models entirely from scratch.

The report specified that Moonshot AI's Kimi K3.1 Pro model, released last week, and DeepSeek's upgraded V4.1-Flash 1017 model, launched on October 17, benefited from distilling the Astra model.

Previously, in a blog post published on September 30, OpenAI stated that it had linked a core part of a distillation campaign starting in July to individuals associated with Moonshot AI.

Major artificial intelligence firms, including OpenAI, Anthropic, and Google, have consistently argued that the practice enables competitors to bypass foundational research and development costs. In response to these concerns, companies have implemented protective measures. Earlier this month, Anthropic reported that new safeguards designed to prevent rivals from copying the capabilities of its Claude AI models successfully reduced data harvested through distillation attacks by approximately 90% to 99%.

Following the publication of the report, OpenAI, DeepSeek, and Moonshot AI did not immediately respond to requests for comment from Reuters.

"The debate around model distillation highlights a critical commercial challenge for foundational AI developers. As competition between US and Chinese firms grows, protecting proprietary model outputs from rapid replication will become a primary focus for venture-backed and enterprise AI companies alike. Founders must anticipate stricter security measures and data-harvesting defenses as the industry matures and regulatory scrutiny increases." — Dr. Shishir Gupta, Founder & CEO, StartupLanes