The OpenAI DeepSeek distillation controversy is now at the center of a growing geopolitical and technological standoff. OpenAI has formally warned US lawmakers that Chinese AI firm DeepSeek is allegedly using advanced “distillation” methods to extract outputs from leading American AI models to strengthen its own chatbot, R1.

According to a memo sent to the House Select Committee on China and reviewed by Bloomberg, OpenAI claims DeepSeek engaged in what it described as “ongoing efforts to free-ride on the capabilities developed by OpenAI and other US frontier labs.” The company says it detected “new, obfuscated methods” designed to bypass safeguards meant to prevent misuse.

What Is OpenAI DeepSeek Distillation?

At the heart of the OpenAI DeepSeek distillation issue is a technical practice known as model distillation. In simple terms, distillation allows one AI system to train itself using outputs generated by another, often more advanced, model.

OpenAI alleges that DeepSeek relied on this approach to replicate or approximate the capabilities of US frontier models, potentially accelerating the development of its R1 chatbot without incurring the same infrastructure costs.

OpenAI reportedly began investigating these practices shortly after DeepSeek released R1 last year. The company even worked with Microsoft to probe whether data had been obtained in an unauthorized way.

Business Threat to US AI Labs

The OpenAI DeepSeek distillation claims are not just about intellectual property—they’re also about competitive positioning.

Unlike OpenAI and Anthropic, which charge subscription fees for premium AI services and have invested billions into AI infrastructure, many Chinese AI models—including DeepSeek—do not require monthly subscriptions. If distillation enables such companies to replicate advanced capabilities cheaply, it could undercut American firms’ business models.

OpenAI’s memo warns that this imbalance could erode the US advantage in artificial intelligence, especially if copied models can scale globally at lower cost.

Lawmakers React to OpenAI DeepSeek Distillation Allegations

The political response has been swift. Representative John Moolenaar, chair of the House China committee, characterized the situation as part of a broader strategy of “steal, copy, and kill,” referring to the Chinese Communist Party.

Meanwhile, former House panel leader Michael McCaul expressed concern about chip export policies, warning that even less powerful Nvidia chips had enabled China to develop highly competitive open-source models.

The OpenAI DeepSeek distillation debate is now tied to broader US-China tensions around semiconductor exports, AI leadership, and national security.

Guardrails, Evasion, and National Security Risks

OpenAI claims that DeepSeek employees may have tried to bypass protections by accessing models through third-party routers and unauthorized resellers. The memo also alleges that programmatic methods were used to retrieve outputs at scale.

Beyond commercial implications, OpenAI flagged national security concerns. It noted that DeepSeek’s chatbot reportedly censors content on topics sensitive to the Chinese government, such as Taiwan and Tiananmen Square. OpenAI warned that when models are copied via distillation, safety safeguards may not transfer—raising the risk of misuse in high-stakes fields like biology or chemistry.

White House AI Czar David Sacks previously suggested there was “substantial evidence” that DeepSeek distilled knowledge from OpenAI models, adding fuel to the OpenAI DeepSeek distillation dispute.

Chips, Exports, and DeepSeek’s Infrastructure

The controversy also intersects with chip policy. Late last year, the US eased certain restrictions, allowing Nvidia to sell H200 processors—about 18 months behind its most advanced Blackwell chips.

DeepSeek’s earlier V3 base model reportedly required 2.8 million H800 GPU hours for training. Those chips were briefly allowed for sale to China in 2023 before new restrictions halted further exports.

US authorities have reportedly examined whether DeepSeek circumvented export controls via Singapore. Records reviewed by the House China committee also show Nvidia provided technical support to help refine DeepSeek’s R1 model.

While DeepSeek has issued only minor updates since R1, Bloomberg reports suggest it is working on an agent-based model to compete directly with OpenAI and other major labs.

The Bigger Picture

The OpenAI DeepSeek distillation conflict underscores a broader shift in the AI arms race. As frontier labs pour billions into training next-generation models, distillation and model extraction tactics could redefine the economics of AI competition.

At stake is more than just code—it’s control over the next wave of intelligent systems, global influence, and the future balance of technological power.

The coming months will reveal whether tighter safeguards, export controls, or diplomatic pressure can contain what is rapidly becoming one of the most consequential AI disputes of the decade.