OpenAI Accuses China's DeepSeek of Distilling U.S. AI Models to Gain Competitive Edge

OpenAI has leveled serious allegations against Chinese artificial intelligence firm DeepSeek, accusing it of systematically distilling the outputs of U.S.-developed large language models—including ChatGPT—to accelerate its own AI development, according to internal documents reviewed by multiple outlets. The claims, first reported by Reuters and corroborated by Rest of World, have ignited a fresh chapter in the escalating technological rivalry between the United States and China in the field of artificial intelligence.

According to Reuters, OpenAI’s internal analysis suggests DeepSeek did not train its models from scratch using raw public data, but instead employed model distillation—a technique where a smaller AI learns to mimic the behavior of a larger, more sophisticated one. This method, while technically legal in many jurisdictions, has raised ethical and strategic concerns when applied to proprietary systems without permission. OpenAI alleges that DeepSeek’s approach allowed it to replicate key reasoning capabilities, multilingual fluency, and coding proficiency typically associated with GPT-4 and other advanced U.S. models, effectively "free-riding" on billions of dollars in American R&D investment, as noted by Rest of World.

While distillation is a well-documented technique in machine learning—used by researchers to create more efficient models—OpenAI argues that DeepSeek’s implementation crosses a line. Internal memos cited by Reuters indicate that DeepSeek’s training data included not only publicly available outputs from ChatGPT and other models but also synthetic data generated by those same models, creating a feedback loop that amplified their performance without direct access to the original training datasets or weights. This practice, OpenAI contends, undermines the value of proprietary research and creates an uneven playing field in global AI markets.

DeepSeek has not publicly responded to the allegations, but prior statements from Chinese AI firms have emphasized their commitment to open-source innovation and independent development. In a 2025 interview with The South China Morning Post, a DeepSeek spokesperson stated the company "builds on public knowledge and open benchmarks," a position that now faces intense scrutiny. Analysts suggest that if proven, OpenAI’s claims could trigger regulatory scrutiny in the U.S. and potentially lead to export control measures targeting AI training tools or data transfers.

The timing of the accusation is significant. DeepSeek is preparing to unveil its latest model, DeepSeek-V3, which industry insiders say rivals GPT-4 in performance metrics while being significantly more cost-efficient to run. This could make it a preferred choice for enterprises in emerging markets and for developers seeking affordable alternatives to U.S.-based models. OpenAI’s public rebuke may be an attempt to deter adoption before the launch, leveraging geopolitical tensions to influence global perception.

Meanwhile, experts warn that such disputes highlight a broader systemic issue: the lack of international norms governing AI model training, data provenance, and intellectual property in the age of generative AI. "We’re in a Wild West of machine learning," said Dr. Lena Zhang, a professor of AI ethics at Tsinghua University, in an interview with Rest of World. "Techniques like distillation exist in a legal gray zone. Without global standards, we’re going to see more of these conflicts—and they’ll only get more complex."

As the U.S. and China continue to diverge in their AI governance frameworks—with Washington pushing for stricter controls and Beijing emphasizing technological self-reliance—the DeepSeek controversy may become a defining case study in the future of AI innovation, competition, and accountability.