OpenAI Levels Serious Allegations Against Chinese AI Rival DeepSeek
In a dramatic escalation of the global artificial intelligence race, OpenAI has formally accused the Chinese AI company DeepSeek of systematic cheating and "free-riding" on American technological advancements. The allegations, detailed in a comprehensive memo to the US House Select Committee on China, paint a picture of sophisticated efforts to replicate and distill frontier AI models developed by OpenAI and other US labs.
The Genesis of a Rivalry and the Accusation of Cheating
The DeepSeek R1 model made its notable debut in January of last year, during the Lunar New Year. At the time, the launch was met with acknowledgment from OpenAI's own CEO, Sam Altman, who publicly called the model "impressive" and expressed that it was "legit invigorating to have a new competitor." However, the tone has shifted radically over the past year.
OpenAI now asserts that DeepSeek is engaged in "adversarial distillation"—a technique where a smaller model is trained using the outputs of a more advanced system. While distillation has legitimate uses, OpenAI alleges that DeepSeek is employing it to unlawfully copy frontier capabilities without the corresponding investment in research, development, or safety infrastructure.
The core of the accusation lies in observed circumvention tactics. OpenAI's memo states that it has identified accounts associated with DeepSeek employees developing methods to bypass OpenAI's access restrictions. These methods reportedly include using obfuscated third-party routers and other techniques designed to mask the source of the queries, allowing DeepSeek to programmatically access and extract outputs from US AI models for training purposes.
Broader Concerns: Censorship, State Support, and the Compute Race
OpenAI's complaint extends beyond mere technical replication. The memo highlights significant concerns regarding DeepSeek's operational alignment with Chinese Communist Party (CCP) policies.
The company alleges that DeepSeek models exhibit severe pro-CCP bias and implement overlapping forms of censorship, not only for users within China but for global users as well. This includes outright refusals to answer questions on sensitive topics like Taiwan or Tiananmen Square, the use of positive language for PRC achievements, and a dynamic censorship layer in the chat interface that can delete generated responses in real-time.
Furthermore, OpenAI underscores the immense state-backed support DeepSeek receives within China's AI ecosystem. This includes large subsidies, preferential procurement, and a national strategy aiming for global AI leadership by 2030. A critical component of this competition is compute capacity—the combination of processing chips and electrical power.
OpenAI points to a growing "electron gap," noting that China added 543 gigawatts of new power capacity in 2025, a figure it claims is ten times the amount added by the United States. The company argues that "infrastructure is destiny" in the AI race, determining which nations can train and deploy the most advanced systems.
OpenAI's Response and Call for Action
In response to these perceived threats, OpenAI details its own multi-pronged strategy. This includes investing in stronger detection systems to identify and prevent unauthorized distillation, banning accounts engaged in such activity, and hardening its models against extraction techniques. The company is also advancing its Stargate Project, aiming to expand US AI infrastructure capacity to 10 gigawatts by 2029.
However, OpenAI advocates for a broader, ecosystem-wide approach. It warns that if only one lab strengthens its defenses, adversaries will simply target the least protected provider. The memo calls for increased US government partnership in areas such as intelligence sharing, establishing industry norms on distillation defenses, and restricting adversary access to US compute and cloud infrastructure.
OpenAI frames the conflict as a fundamental contest between democratic and autocratic models of AI development. It positions its own work—making powerful tools broadly available with safety guardrails—as the embodiment of "democratic AI," in contrast to systems it says are built for censorship and state control. The company concludes by expressing confidence that open markets and responsible innovation will sustain democratic leadership, but emphasizes the urgent need for strategic investment and vigilance.
