Chinese Firms DeepSeek and Moonshot AI Train New AI Models Using GPT-6 Astra Amid OpenAI’s Distillation Crackdown

Distillation Techniques in AI Development

According to НВ — Техно: Chinese tech companies DeepSeek and Moonshot AI have harnessed responses from OpenAI’s GPT-6 Astra model to develop their latest artificial intelligence systems through a process known as model distillation. Moonshot AI launched its Kimi K3.1 Pro model at the end of October 2026, while DeepSeek released an upgraded version, DeepSeek V4.1-Flash 1017, on October 17, 2026.

Model distillation involves training a smaller or less complex AI system by leveraging outputs from a more advanced model. Although this method can accelerate AI development, it has raised concerns among leading AI organizations-including OpenAI, Anthropic, and Google-due to the risk of replicating cutting-edge capabilities without proper authorization. OpenAI announced on September 30, 2026, that it had uncovered a distillation campaign linked to Moonshot AI, which reportedly began in July 2026.

Ethical and Security Challenges

Meanwhile, Anthropic reported successfully reducing data exposure from model-copying attacks against its Claude AI by 90–99%, thanks to newly implemented security measures. Neither OpenAI nor the Chinese companies DeepSeek and Moonshot AI have publicly commented on these developments. This episode highlights the growing need for vigilant monitoring and regulation within the AI sector as developers explore innovative training methods.

The situation underscores the critical importance of ethics and security in AI innovation. While distillation can drive progress, it also poses potential threats to intellectual property rights and fair competition. As concerns over unauthorized technology replication increase, it becomes essential for regulators and industry leaders to establish robust protections to safeguard innovation while preventing misuse and ensuring a level playing field in AI advancement.

As the advancements in AI continue to unfold, the recent enhancements in DeepSeek's latest model raise critical questions about the balance between innovation and security. While DeepSeek V4.1-Flash 1017 boasts improved safety features, it also presents new risks due to compromised filters, highlighting the ongoing challenges in ensuring robust protections in AI development. This situation underscores the importance of monitoring ethical standards as the industry evolves.