Home · Technology · Sep 8 archive

U.S. Agencies Warn Chinese AI Companies of Systematic Model Copying

Developing

Technology Desk

In Short: U.S. agencies have warned that top Chinese AI companies have been systematically copying American models, raising concerns about intellectual property and safety.

U.S. agencies have warned that top Chinese AI companies have been systematically copying American models, according to reports. The FBI, NSA, and CISA issued an alert on Tuesday, stating that companies like Anthropic, OpenAI, Google, and xAI have been mined by Chinese counterparts since 2024.

The report, which was published by NBC News, indicates that the scale and sophistication of these campaigns suggest that distillation is not a supplementary activity but a critical core of the Chinese companies' AI model development. Distillation, a process where smaller, more efficient models are created from larger ones, is widespread in the AI industry, though it often violates terms of service.

According to reports, the practice has been ongoing for years. OpenAI, for instance, accused DeepSeek of 'inappropriately' basing its model on ChatGPT in 2023. The U.S. government has increasingly alleged that Chinese models are built by distilling American ones, with the practice being against AI companies' terms of service.

The U.S. agencies' report highlights the potential risks of this practice, including the circumvention of safety guardrails built into American systems. President Donald Trump is scheduled to meet with Chinese President Xi Jinping on September 15, which could potentially address these concerns.

Traditional weather models, which rely on reanalyses, have been compared to AI weather models. While AI models can achieve similar forecast performance with less computing power, they often require estimates for conditions without real-world measurements. This highlights the complexity and challenges in developing robust AI models.

What's still developing

Sources