Signal

Former OpenAI researcher says he found Nvidia Blackwell fingerprints hidden inside six Chinese AI models

First reported by The Decoder ·

The signal ●●○○ Compiled by AI from The Decoder, Wccftech, WinBuzzer and Business Today
Why you might care

US-controlled advanced AI hardware is likely being used by Chinese firms, potentially reshaping the global AI development race.

What happened

AI researcher Will DePue claims to have found the first technical evidence that major Chinese AI companies have access to Nvidia's latest Blackwell GPUs, which are subject to US export controls. DePue's analysis, verified by senior researchers at OpenAI and Anthropic, suggests that models from DeepSeek, Moonshot AI, MiniMax, and Ant Group released since April 2026 bear the "fingerprint" of Blackwell hardware, distinct from older Hopper GPUs. The method analyzes statistical patterns in the initial weights of AI models, which vary based on the GPU's core count during the seeding process. DePue's findings indicate that specific Chinese models likely used Nvidia's 148-core B200/B300 or 152-core GB200/GB300 chips, with training runs estimated to cost hundreds of millions of dollars. While this evidence points to access to restricted hardware, it does not definitively prove illegal smuggling or violations of US export law, as companies could be utilizing cloud computing services outside China.

What it means

The technical method used by DePue is groundbreaking, as it bypasses the difficulty of directly monitoring hardware usage and instead infers it from the digital artifacts left within trained AI models. This "fingerprinting" technique could become a critical tool for understanding global AI hardware distribution and identifying potential violations of export controls. Its validation by researchers at leading AI labs like OpenAI and Anthropic lends significant credibility to the findings, suggesting that this is not an isolated observation but a potentially widespread phenomenon.

If these findings hold, they imply that Chinese AI companies are accelerating their development by leveraging cutting-edge hardware that was intended to be restricted. This could lead to a more intense competitive landscape, with implications for national security and the global balance of AI power. It also raises questions about the effectiveness of current export control measures and may prompt stricter enforcement or new strategies to prevent the circumvention of these regulations.

AI-written summary. May contain errors.

AI Chips