OpenAI tells Congress China’s DeepSeek bypassed key access safeguards

OpenAI has warned the US government that Chinese artificial intelligence startup DeepSeek is attempting to replicate American AI systems by bypassing platform safeguards and extracting model outputs.

In a memo sent Thursday to the US House Select Committee on Strategic Competition between America and the Chinese Communist Party, the developer of the large language model ChatGPT accused its Hangzhou-based competitor of systematically ripping off technology developed by US frontier labs.

The report claims DeepSeek has been “free-riding on the capabilities developed by OpenAI and other U.S. frontier labs,” a practice known as distillation.

“We have observed accounts associated with DeepSeek employees developing methods to circumvent OpenAI’s access restrictions and access models through obfuscated third-party routers and other ways that mask their source,” the company wrote. “DeepSeek employees developed code to access US AI models and obtain outputs for distillation in programmatic ways.”

Washington and Beijing have been competing for the top AI spot since DeepSeek debuted about 12 months ago. After the Chinese LLM model was released, the US House Select Committee observed that the CCP requested OpenAI to investigate whether DeepSeek had used any US tech or AI chips in its development.

ChatGPT maker accuses DeepSeek of illegal distillation

According to the Sam Altman-led tech firm, Chinese actors are using information pipelines to mimic the methods of US AI synthetic data generation labs. The company also reported that some Chinese firms have created networks of unauthorized resellers of OpenAI services to evade law enforcement.

OpenAI also cautioned that copying capabilities through adversarial distillation, without equivalent safety frameworks, may produce systems that lack consumer protections, albeit cheaper to scale. It said shortcomings in such systems might only surface after deployment, when risks are harder to manage.

Beyond technical allegations, OpenAI’s memo noted that DeepSeek’s content governance was found to be politically biased and to impose extensive censorship. Within the company’s purview, the most widely used LLM in China showed a severe pro-CCP bias in recent releases.

OpenAI said that when DeepSeek was asked questions on topics sensitive to Beijing, such as Tiananmen Square or Taiwan independence, it frequently issued outright refusals. In other cases, DeepSeek issued biased responses to PRC-favored stories and redirected prompts that appeared to be criticism of the CCP.

US has an advantage over China due to tech chips

According to OpenAI’s memo to US policymakers, the scarcest resource in AI is the combined power and chip resources required to execute code, defined as compute. It said that sustaining the American advantage depends on the ability to generate and deliver electricity at scale to support computational demands.

Last month, two sources familiar with the matter told reporters that Chinese authorities have approved DeepSeek to purchase Nvidia’s H200 artificial intelligence chips, subject to regulatory conditions that are still being finalized. 

US President Trump greenlighted Nvidia’s request to ship H200 chips to Beijing in early January, but Chinese regulators have the final authority to permit the shipments. At the time, Nvidia Chief Executive Jensen Huang said his company had not received word of China’s approval.

Earn 8% CASHBACK in USDC when you pay with COCA. Order your FREE card.