OpenAI Warns US Lawmakers That Chinese AI Firm DeepSeek is Copying Outputs to Train Rival Chatbots

DeepSeek, the Chinese artificial intelligence firm that recently grabbed global attention, is now facing heavy accusations from OpenAI.

In a memo sent on 12 February 2026 to the US House Select Committee on China, the Sam Altman-led company claimed that DeepSeek has been using increasingly clever ways to extract data from American models to train its own systems.

Digital Shortcuts and the Art of Distillation

At the heart of the dispute is a training method called distillation.

This is where a newer, smaller AI model learns by watching how a more advanced model, like ChatGPT, answers questions.

Instead of building from scratch, the new model effectively mirrors the logic and patterns of the "teacher" system.

OpenAI alleges that DeepSeek is "free-riding" on billions of dollars of American investment.

They claim that accounts linked to DeepSeek employees have been using "programmatic ways" to pull massive amounts of data.

To stay hidden, these accounts reportedly used third-party routers and "obfuscated methods" to mask where the requests were coming from.

Representative John Moolenaar, chair of the House committee, said,

“Chinese companies will continue to distill and exploit American AI models to their advantage, just like when they ripped off OpenAI to build DeepSeek.”

Why the US is Concerned About This Practice

While distillation is a known technical concept, OpenAI argues that the scale and secrecy of DeepSeek’s approach is the problem.

Because DeepSeek offers its models for free, while US companies like OpenAI and Anthropic charge subscription fees to recover their costs, there is a fear that the US competitive lead is being chipped away.

Beyond business, there are safety worries.

OpenAI told lawmakers that when a model is copied this way, the safety guardrails often don't carry over.

This could lead to a version of the AI that helps with dangerous tasks in biology or chemistry without the usual restrictions.

There is also the issue of political filtering.

Internal reviews suggested DeepSeek’s chatbot would avoid or censor topics sensitive to the Chinese government, such as the Tiananmen Square protests.

Hardware Loopholes and the Race for Power

The software battle is only one half of the story.

There is a secondary row over the physical chips needed to run these massive systems.

At the end of 2025, the US government moved to allow Nvidia to sell its H200 processors to China, despite those chips being about 18 months behind the latest Blackwell technology.

DeepSeek-V3 required 2.8 million H800 GPU hours for its full training, a figure that highlights the massive infrastructure already in place.

Recent records show that Nvidia actually provided technical support to help DeepSeek co-design the R1 model.

This has sparked anger in Washington.

Representative Michael McCaul expressed his frustration, noting that China managed to build world-class models using older, less powerful chips.

McCaul added,

“DeepSeek should have been a wake-up call about the dangers of selling advanced semiconductor chips to the CCP.”

Investigations Stretching from Washington to Singapore

The pressure isn't just coming from politicians.

US authorities have opened a probe into whether DeepSeek bypassed export controls by buying chips through intermediaries in Singapore.

This follows reports that while Singapore accounts for a huge chunk of Nvidia's revenue, very few of those chips actually stay on the island, raising questions about where they eventually end up.

For now, DeepSeek has not responded to the latest memo, but the company continues to push ahead.

While it has only released minor updates since the R1 launch, it is reportedly working on an agent-based model to take on its American rivals directly.