On 2 July 2026, the China National Intellectual Property Administration published detailed analyses of the ten representative patent reexamination and invalidation cases for 2025, expanding on the case summaries released in April. In the AI-related case concerning the generation of dynamic images from audio, the patent was upheld and the panel treated model design, model training and model application as distinct stages when applying the inventive-step test. The analysis indicates that examiners should look at the application scenario, data or object being processed, the training method and the resulting technical effect, rather than dismissing a new use of an algorithm as a routine substitution.
The pharmaceutical case involving Pfizer’s crizotinib crystal form also sharpened the standard for post-filed experimental data. Such evidence is not automatically excluded, but the claimed technical effect must be derivable from the original disclosure, the testing method must be sound, the results reliable, and the evidence open to challenge by the other party. The practical message is straightforward: in complex technologies, patent strength increasingly depends on what the original filing actually explains about technical effects, training or testing conditions, and the basis for comparison. A traceable evidentiary record at filing is far more dependable than a broad explanation added later.
