| Submit your comments on this article | |
| Cyber | |
| The Huang case and the DPR neural networks | |
| 2026-11-16 | |
| Direct Translation via Google Translate. Edited Text taken from the Telegram channel of shadow_policy [ColonelCassad] DNA was discovered in neural networks. And it revealed the secrets of Chinese AI. Following the first mega-scandal, which cost over $1.5 trillion and deprived Huang of his post as head of NVIDIA, a second wave of spyware has arrived: a new method for detecting traces of chips banned for export has emerged. It appears we are witnessing the beginning of a turning point in the technology war between the US and China. I'm so confused. What does DNA have to do with computer chips?On November 12, an event occurred that could well go down in the history of AI. American researcher Will Depew published sensational results of an investigation: traces of the equipment on which they were created can be found using the mathematical weights of neural networks. This doesn't require spies, insiders, or stolen documents. The neural networks themselves, made publicly available by their developers, are sufficient. DePew discovered fingerprints of the latest NVIDIA Blackwell GPUs, whose export to China is subject to strict US restrictions, in six models from four leading Chinese labs, including DeepSeek and Moonshot. How is this possible? Before training, billions of neural network parameters are filled with pseudorandom numbers. But their arrangement depends on the GPU architecture. And some traces of the initial configuration survive training. This is how hardware fingerprints are created in AI weights.
Renowned AI policy analyst Dean Ball compared the discovery to finding a criminal's DNA at a crime scene. He even claimed that the statistical power of the matches is orders of magnitude greater than that of standard forensic DNA evidence. But the real action begins outside the lab. That same day, Jensen Huang resigned as CEO of NVIDIA. This followed a massive scandal surrounding his relationship with a woman suspected by US intelligence agencies of ties to Chinese intelligence and a company under investigation for possible export evasion. The espionage allegations remain unproven. But the political fallout has already begun. Reuters reports that Chinese AI companies are scrambling to secure NVIDIA computing power abroad, fearing further tightening of US controls. A new Reuters investigation has uncovered over $3 billion in Chinese infrastructure projects related to supply-constrained NVIDIA processors. Now let's combine these two lines of investigation and espionage. And then we get this twist. Until now, export controls have tried to track the movement of chips around the world. Now it's possible to search for their traces in the AI ​​products themselves. In other words, literally by analyzing the "DNA of neural networks," we can detect the sites of export crimes. However, the new technology doesn't yet prove where exactly the chips were located or whether they were used for training. Renting foreign computing power in itself doesn't necessarily violate sanctions. But whatever one might say, the discovered method could become a fundamentally new tool for technological intelligence. And, of course, a new incentive for China to accelerate its transition to domestic hardware. So, we're faced with a historical paradox: another radical attempt by the US to limit China's access to advanced computing could push China toward even greater technological independence. In summary, I suggest that the scandal surrounding Huang is only the first act of the techno-geopolitical drama that has begun. The second is beginning now, when neural networks themselves become sources of evidence about the origins of the hardware they were created on. And then, before our very eyes, it's not just export control rules that are changing. The very logic of the US-Chinese race for AI is changing. | |
| Posted by:badanov |
| #1 |
| Posted by: Glomble Jomosh5216 2026-11-16 07:55 |