On 8 September 2026, the China National Intellectual Property Administration (CNIPA) issued implementation guidance aimed at improving the efficiency and coordination of patent and trademark examination in areas linked to artificial intelligence. For patent applications covering large models, algorithm architectures and inventions combining software and hardware, the guidance places clearer emphasis on alignment with data-security compliance and the lawful provenance of training data. The practical effect is that applicants may need to ensure that patent drafting, R&D records and internal data-governance documentation tell a consistent story, especially where the claimed technical contribution depends heavily on the collection, preparation or use of training datasets.
On the trademark side, the guidance further develops evidentiary standards in non-use cancellation proceedings, with particular attention to electronic evidence and attempts to manufacture a record of use through sham invoices, nominal transactions or defensive arrangements. The direction of travel is clear: isolated documents are likely to carry less persuasive weight than a traceable chain of genuine commercial use linking contracts, orders, payment records, logistics, online listings and promotional activity. For rights holders, that raises the value of disciplined evidence retention; for challengers, it creates more room to test whether the asserted use reflects real market activity rather than paperwork assembled to defeat cancellation.
