OpenAI has formally accused Chinese AI startup DeepSeek of building its advanced artificial intelligence systems by copying American technology. In a memo sent to the U.S. House Select Committee on China on February 12, the ChatGPT maker claimed DeepSeek is using a technique known as “distillation” to “free ride” on the capabilities developed by U.S. research labs.
The San Francisco-based company alleges that DeepSeek employees have been using “new, obfuscated methods” to bypass security and harvest data from OpenAI’s servers. According to the memo, this allows the Chinese firm to create the illusion of parity with American models without investing the billions of dollars typically required for such breakthroughs.
The ‘Distillation’ Accusation
OpenAI’s primary complaint centers on “distillation,” a process where a smaller, cheaper AI model learns by mimicking the answers of a larger, more powerful one. Instead of learning from raw data—which is expensive and computationally intensive—the smaller model simply treats the advanced model like a teacher and copies its logic.
While distillation is a common industry practice for optimizing software, OpenAI argues that DeepSeek is using it to compete unfairly. By “extracting results” from leading U.S. models, DeepSeek can allegedly train its own systems at a fraction of the cost. OpenAI stated that this practice risks eroding the American advantage in artificial intelligence and creates a misleading picture of China’s technological progress.
‘Obfuscated Methods’ to Evade Detection
The memo details specific tactics allegedly used by DeepSeek to evade OpenAI’s defenses. OpenAI claims it detected accounts associated with DeepSeek employees using third-party routers to mask their location and identity.
Furthermore, the company says DeepSeek developers wrote code to access U.S. models in “programmatic ways,” allowing them to automate the collection of training data. OpenAI noted that despite its efforts to block unauthorized users, “networks of unauthorized resellers” have helped actors circumvent these controls. The company is now proactively removing users who appear to be engaging in this data scraping.
Lawmakers React: ‘Steal, Copy, and Kill’
The accusations have drawn sharp responses from U.S. officials concerned about national security. Representative John Moolenaar, referencing the findings, described the alleged strategy as a standard operation for Chinese competitors.
“This is part of the CCP’s playbook: steal, copy, and kill,” Moolenaar said, warning that Chinese companies will continue to exploit American innovation to their advantage.
Representative Michael McCaul also weighed in, stating that DeepSeek’s rapid rise should serve as a “wake-up call” regarding the sale of advanced hardware to China. He expressed concern about what foreign firms might achieve if they gain access to even more powerful tools, such as Nvidia’s H200 chips.
Market Context and Nvidia’s Role
The controversy comes roughly a year after DeepSeek released its R1 model, which shook global markets by offering high-level performance at a surprisingly low price. DeepSeek claimed it trained the model for just $6 million, a figure that baffled many Western experts. OpenAI’s memo suggests this efficiency was largely due to distillation rather than a novel breakthrough.
Documents obtained by the House committee also revealed that U.S. chipmaker Nvidia provided technical support to help DeepSeek design and improve its R1 model. According to reports, DeepSeek utilized Nvidia’s H800 processors—chips that were briefly allowed for export to China before restrictions were tightened—to train its system.
