Six Chinese AI companies have been accused of aggressively and systematically copying US frontier AI models, according to a new report that highlights the growing intellectual property tensions in the global AI race. The allegations suggest that despite export controls and other restrictions, Chinese firms have found ways to replicate the capabilities of leading American AI systems.
The report details patterns of behavior that go beyond normal competitive benchmarking. Accused firms allegedly used techniques including distillation, training smaller models on the outputs of larger ones, and in some cases, direct extraction of model weights through various technical means. These practices, while not always illegal, raise serious questions about the boundaries of legitimate competition.
US frontier labs have invested billions in developing their models, and the copying allegations threaten to undermine the economic incentives for continued investment. If competitors can simply replicate frontier capabilities at a fraction of the cost, the rationale for massive R&D spending weakens significantly.
The accusations come at a sensitive time for US-China tech relations. Export controls on advanced AI chips and model weights were designed to slow China AI progress, but the report suggests these measures have been less effective than hoped. Chinese firms have adapted by developing workarounds and alternative technical approaches.
The situation underscores a fundamental challenge in AI governance: how do you protect intellectual property in a field where the product is knowledge encoded in neural network weights, and where the techniques for extracting that knowledge are themselves published in academic papers?
