Guangdong's count of registered large models reaching 195 was formally disclosed on October 16, 2026, at the results conference of the Greater Bay Area Generative AI Safety Development Joint Laboratory in Guangzhou. The Guangdong Registered Large Model Observation Report (2026) released at the event shows that since registration work began in 2023, the province's registered generative AI services have grown from 8 to 195, a 23-fold increase in three years, accounting for nearly one fifth of the national total as of September, with registration volume, industry coverage and deployment scale all among the country's highest. A new batch of 31 Guangdong companies received their registration certificates at the event.
Behind the volume, testing and governance are being put on the schedule
The 195 figure is only a stock number; what matters more is the supporting system announced around it. The laboratory's Technical System Construction Plan (2026-2028), released at the same event, says that as AI moves from large models toward agents and scaled deployment, the lab will focus on safety testing, situational awareness, agent and application safety assessment, trustworthiness evaluation, cross-border governance coordination within the Greater Bay Area and AI safety think-tank work. Since its inauguration in September 2025, the lab has deployed a locally hosted generative AI safety evaluation system and built a test bank at million-question scale, covering corporate self-assessment, registration verification and post-launch dynamic inspection.
The APEC AI Legal Policy Blue Book (2026), also released at the event, looks beyond the region with a comparative study of AI law and policy across all 21 APEC economies. The conference also inaugurated the lab's Guangzhou service station and Shantou service center, pushing post-registration services down to the city level. For companies, the practical effect is that registration is no longer just a launch permit: testing, inspection and cross-border compliance advice now have nearby institutions to handle them, which can spread compliance costs more thinly.
The real impact for the industry comes after registration
Leading registration numbers do not by themselves mean leading model capability; the signal from Guangdong is that governance infrastructure is being built to match industrial scale. Once registrations grow from 8 to 195, the regulatory question shifts from whether a service may launch to whether it stays compliant afterward, and how agent applications are assessed and held accountable when something goes wrong. As our earlier report on the proposal to add 23 AI technologies to the export control list showed, rules on the export side are tightening at the same time. Expanding registrations on one side and stronger testing, inspection and outbound compliance on the other will become the normal shape of regional large-model competition. For teams preparing to launch generative AI services in Guangdong, the pragmatic move now is to schedule safety testing and registration materials as part of the product process early, rather than patching them in just before launch.
