A researcher finds forensic proof that Chinese AIs are secretly using Nvidia's banned chips.
Original headline: Former OpenAI researcher says he found Nvidia Blackwell fingerprints hidden inside six Chinese AI models — the-decoder.com
More in AI and machine learning news
Former OpenAI researcher Will DePue has published a forensic analysis claiming to have found the first technical evidence that several top Chinese AI labs are using Nvidia's latest, export-restricted Blackwell GPUs. The report names DeepSeek, Moonshot AI, MiniMax, and Ant Group, suggesting their models released since April 2026 carry a distinct Blackwell signature, while older models show traces of the previous Hopper generation. DePue says his findings have been independently checked by senior researchers at OpenAI and Anthropic.
DePue's method relies on a telltale pattern left in a model's weights. The PyTorch framework initializes these weights with random numbers, a process split across a GPU's compute cores. Because Blackwell chips have a different number of cores than older Hopper chips, they leave a unique statistical fingerprint, which DePue claims can be detected even after extensive training. This analysis, if confirmed, would be a major forensic breakthrough and a significant breach of US export controls, revealing how cutting-edge AI development is still being powered by precisely the chips the regulations intended to block.
Read the original at the-decoder.com
Satirical commentary on real tech news, summarised with AI. Read the original source for the facts. How ZeroDaily works
