Key Highlights:

• OpenAI claims DeepSeek used AI distillation to replicate US models.
• Memo alleges efforts to bypass access controls using masked routing.
• Lawmakers warned as China’s AI progress raises security concerns.
• DeepSeek and parent firm High-Flyer have not responded.

OpenAI has warned US lawmakers that China-based AI startup DeepSeek allegedly trained its models by distilling American AI systems. The disclosure matters because it raises new concerns over AI security, fair competition, and how easily advanced models can be replicated across borders.

The warning appears in a memo sent to the U.S. House Select Committee on Strategic Competition, reviewed by Reuters.

What exactly is OpenAI accusing DeepSeek of?

In the memo, OpenAI claims that DeepSeek employees actively sought to access US AI models, including OpenAI systems, to extract outputs for training their own models.

According to OpenAI, some accounts linked to DeepSeek allegedly bypassed access controls by routing requests through obfuscated third-party services. These methods, OpenAI says, helped mask the source of the requests.

The memo also states that DeepSeek developers built programmatic tools to collect AI-generated outputs at scale. These outputs were then used for AI distillation, a process that transfers knowledge from stronger models to weaker ones.

What is AI distillation, and why is it controversial?

AI distillation involves using a powerful, established model to evaluate and refine the responses of a newer model. Over time, this allows the newer system to mimic advanced reasoning without full access to original training data.

While distillation is common in AI research, it becomes controversial when companies allegedly extract outputs without permission. OpenAI describes this behavior as an attempt to “free-ride” on US-developed AI capabilities.

Why is DeepSeek already on Washington’s radar?

Hangzhou-based DeepSeek gained global attention early last year after releasing models that rival top US offerings. Models like DeepSeek-V3 and DeepSeek-R1 earned praise from Silicon Valley engineers and became widely available.

However, the speed of DeepSeek’s progress has also fueled fears in Washington. Despite US restrictions on advanced chips and AI exports, Chinese AI firms continue to close the gap.

OpenAI claims that several Chinese large language models are cutting corners on safety and deployment standards.

How is OpenAI responding now?

OpenAI says it actively removes accounts suspected of attempting AI distillation for rival model development. The company also frames the issue as a national competitiveness challenge, not just a business dispute.

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

What happens next?

As DeepSeek remains a trending keyword, the memo adds fuel to an already tense AI race between the US and China. Lawmakers may now push for stricter access controls, clearer rules on model usage, and stronger safeguards around AI distillation.

The DeepSeek debate shows how quickly technical practices can turn into geopolitical flashpoints.