OpenAI warns US lawmakers over DeepSeek ‘copying’ AI models: What we know

OpenAI has warned US lawmakers that Chinese artificial intelligence startup DeepSeek is trying to copy advanced US-made AI models and use them to train its own systems.

According to a memo reviewed by Reuters, OpenAI said DeepSeek is attempting to “free-ride” on the work done by OpenAI and other leading American AI labs. The warning was sent to the US House Select Committee on Strategic Competition, which focuses on US-China relations.

What OpenAI told US lawmakers

In the memo, OpenAI said it has seen signs that people linked to DeepSeek tried to bypass OpenAI’s safeguards. According to the company, these efforts included using masked or indirect routes to access its AI models.

OpenAI also claimed that DeepSeek employees wrote code to collect large volumes of AI-generated answers in an automated way. These outputs were then allegedly used to train DeepSeek’s own models.

The company said it regularly removes users who appear to be copying its models in this way.

What is AI ‘distillation’ and why it matter?

The method at the centre of the dispute is known as distillation. Basically, it means using a powerful, well-trained AI model to judge or guide the responses of a newer model.

By repeatedly learning from the stronger model’s answers, the newer system can improve quickly without needing the same level of computing power or original training data.

OpenAI told lawmakers that this technique, when done without permission, allows rivals to benefit from years of costly research without following the same safety or access rules.

DeepSeek has drawn global attention

DeepSeek, based in Hangzhou, attracted global notice last year after releasing AI models that performed close to top US systems. Its models, including DeepSeek-V3 and DeepSeek-R1, were praised by some technology experts for their speed and reasoning abilities.

This progress raised concerns in Washington about how fast Chinese companies could narrow the AI gap, even as the US places limits on advanced chips and technology exports to China.

DeepSeek and its parent company, High-Flyer, did not respond to requests for comment, according to Reuters.

Safety concerns and the wider AI race

OpenAI also told US lawmakers that some Chinese AI developers are cutting corners on safety when training and deploying large language models. It did not provide detailed examples but said this could increase long-term risks.

The warning highlights growing tension in the global AI race, where governments are closely watching how advanced models are built, shared and protected. As AI becomes more powerful, questions around access, copying and regulation are likely to become more important in US-China technology relations.