OpenAI says China's DeepSeek trained its AI by distilling US models, memo shows
A leaked memo reveals explosive allegations of AI model theft and espionage.
Deep Dive
OpenAI has reportedly warned U.S. lawmakers that Chinese AI firm DeepSeek trained its R1 chatbot by secretly distilling data from U.S. models like GPT-4. The memo alleges DeepSeek used obfuscated servers to bypass access restrictions, effectively free-riding on American innovation. This accusation, submitted to the House Select Committee, escalates the tech cold war and raises major intellectual property and national security concerns between the two superpowers.
Why It Matters
This accusation could trigger major regulatory crackdowns and reshape the global AI arms race.