OpenAI accuses the Chinese company DeepSeek of illegally using data to train the R1 model

Brief on the conflict between OpenAI and Chinese startup DeepSeek

When Accusations – that DeepSeek “steals” data from American AI models to train its chatbot R1.OpenAI (creators of ChatGPT) 2024 report addressed to members of US Congress Bloomberg/Source in OpenAI Yesterday Distillation method – allows quickly acquiring knowledge from an already trained model without direct access to the source code. DeepSeek (and other Chinese companies) Since early 2023 Risks – obtained models may carry restrictions set by original developers and be used for malicious purposes.OpenAI/Microsoft Summer 2023 Possible access to Nvidia H200 accelerators – which would boost DeepSeek’s competitiveness in the global AI market US policymakers Since early 2024.

What happened
1. OpenAI accusations

OpenAI, in a statement published by Bloomberg, claims that Chinese startup DeepSeek uses distillation to “freely” transfer technologies developed by American companies, including ChatGPT.

2. Report to American legislators

The document contains a detailed analysis of how DeepSeek bypasses existing protective mechanisms and gains access to data from American AI models through “circumvention” channels. The report is addressed to a special committee of Congress.

3. Why it matters

- Distillation speeds up training of third‑party models without needing source code.
- Obtained models often lack the restrictions applied to original systems and can be used maliciously.
- DeepSeek does not offer paid subscriptions, making its solutions more accessible to a broad audience in China (and even Russia) than commercial Western counterparts.

4. Potential risk from Nvidia H200

US policymakers are concerned that DeepSeek could gain access to the latest Nvidia H200 accelerators after President Trump allowed shipments to China. This could significantly boost the Chinese startup’s technological power and weaken U.S. positions in the global AI arena.

5. Industry reaction

OpenAI and Microsoft have already begun investigating DeepSeek’s activities last year, but so far have not been able to fully close “transitions” through intermediaries that allow the Chinese startup to use American models.

Conclusion
OpenAI claims that its Chinese competitor uses distillation methods for accelerated training and access to protected data. This threatens not only U.S. commercial interests but also security, as obtained models can bypass built‑in restrictions. In response to these threats, U.S. policymakers are considering tightening controls on the export of high‑technology equipment to China.