OpenAI controls DeepSeek "free-riding"! Stealing American AI technology to train their own models

The US-China artificial intelligence competition continues to heat up. Recently, OpenAI submitted a memorandum to the U.S. Congress accusing Chinese AI company DeepSeek of improperly extracting outputs from top American models using “distillation” techniques to train their own R1 chatbot. OpenAI claims this behavior is akin to “free-riding,” threatening the commercial interests of American companies and potentially impacting national security and technological privacy.

OpenAI: DeepSeek Uses “Distillation Technology” to Bypass Defenses

Bloomberg reports that in the memorandum submitted to the House Select Committee on the CCP, OpenAI states that DeepSeek utilizes so-called “distillation” techniques to learn from the outputs of cutting-edge American AI models, thereby enhancing their R1 chatbot capabilities.

Distillation is a common training method in AI, where smaller models learn from the outputs of larger models to improve performance. However, OpenAI claims that DeepSeek employs “a new and obfuscated method” to circumvent platform protections, including hiding sources through third-party routers, accessing services via unauthorized resellers, and programmatically extracting large amounts of output data from American models.

(Can OpenAI turn ChatGPT into a cash cow before burning through hundreds of billions of dollars?)

Impact of Free Models on the US AI Commercial Business Model

OpenAI warns in the memorandum that if distillation practices continue to spread, they could weaken the competitive advantage of American AI companies. Companies like OpenAI and Anthropic have invested billions of dollars in building computing infrastructure and charge subscription or enterprise service fees.

In contrast, DeepSeek and most Chinese models adopt lower or nearly free pricing models. If they can replicate cutting-edge capabilities at lower costs through distillation, it could alter the competitive landscape of AI business models, further shrinking the market share of American firms.

National Security and Chip Export Issues Come to the Fore

Beyond commercial concerns, OpenAI also raises national security issues in the memorandum, including content censorship by DeepSeek’s chatbot on topics like Taiwan and Tiananmen Square, as well as potential weakening of security protections during the distillation process, making high-risk fields such as biological or chemical applications more vulnerable to misuse.

U.S. policymakers have responded strongly. Committee Chair John Moolenaar stated that this is a typical Chinese “theft, plagiarism, and suppression” strategy.

Additionally, Nvidia’s previous sale of H800 chips to China has become a focal point. The U.S. has investigated whether DeepSeek obtained related hardware resources through third parties, intertwining this technological dispute with semiconductor export controls.

(DeepSeek’s One-Year Anniversary of Popularity: How Is China’s AI Model Development Going?)

AI Race Turns Intense

Michael McCaul, former head of the House Export Controls Oversight Panel, commented, “DeepSeek should serve as a warning, reminding us of the dangers of selling advanced semiconductor chips to the CCP.”

China has developed some of the world’s most advanced open-source models using less powerful Nvidia chips. The thought of what they could achieve with more advanced hardware like the H200 chips makes me shudder.

It’s clear that this controversy is no longer just a technical dispute between individual companies but a reflection of the broader AI industry competition and geopolitical struggle. As generative AI technology accelerates, the contest between the U.S. and China over model capabilities, chip supply, and regulatory frameworks is likely to intensify.

This article, “OpenAI Accuses DeepSeek of ‘Free-Riding’! Stealing US AI Technology to Train Its Own Models,” first appeared on Chain News ABMedia.