Washington accuses Chinese companies of copying US AI models
The allegations made by the NSA, CISAs and the FBI: they are reportedly siphoning off the advanced capabilities of US systems
The standoff between China and the US over artificial intelligence continues. In the race for technological supremacy between the two countries, leading Chinese artificial intelligence research centres are accused of ‘systematic’ theft of US technology.
The National Security Agency (NSA), the Cybersecurity and Infrastructure Security Agency (CISA) and the Federal Bureau of Investigation (FBI) have in fact drawn up a cybersecurity warning document, which investigates the activities of certain Chinese companies in relation to the systematic theft of features and capabilities from American models.
The allegations: copied features and distillation techniques
Such practices – the document explains – take place ‘through campaigns such as knowledge distillation on an industrial scale, which form the core – and not merely a complement – of their (the Chinese industries’, ed.) AI development strategy’.
Distillation, in itself, is a widely used machine learning technique in the sector that enables the knowledge of a large, pre-trained model – the ‘teacher model’ – to be transferred to a smaller ‘student model’. Chinese companies, however, are reportedly making ‘malicious’ and ‘aggressively’, benefiting from ‘significantly shorter AI development times and reduced financial costs for training state-of-the-art models’, with the aim of ‘bridging the technological and performance gap’ between the two countries’ IT capabilities.
The companies involved
The joint US investigation names the following: DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI are reportedly the Chinese companies responsible for the systematic theft.
