A document may be generated in seconds, but three questions remain: Is the sourcing reliable? Did the data leave the domain? Was the action authorized? China's State Administration for Market Regulation has now approved GB/T 48324-2026, "Artificial Intelligence — Technical Requirements for Government Large Model Systems" — the country's first recommended national standard for government-sector LLMs, giving civic AI builds a unified technical baseline.
Four lines drawn by the standard
The standard sets systematic requirements across model capability, scenario application, security and compliance, and deployment operations. Kingsoft Office participated in drafting, authoring the government-office scenario modules, after previously leading the office-LLM national standard. "The premise of useful government AI is security and compliance," said Kingsoft VP Wang Shaokang — with humans keeping final say on key steps, civic AI can become "usable, practical and trusted."
Frontline results: 3-hour drafts cut to 40 minutes
Deployment samples
· Zhongshan: 34,000 civil servants equipped with AI drafting assistants; 600,000 documents form a dedicated knowledge base
· Document drafting time down from 3 hours to 40 minutes; content error rate down 95%
· All data flows stay inside the government intranet, fully logged and encrypted on export
· Guangdong: full-stack localized deployment on xinchuang infrastructure, reusing each agency's permission structure
Security is baked into workflows: WPS Comate follows "human-machine equal rights," accessing files strictly per user permissions and requiring re-confirmation for high-risk operations; client and cloud sandboxes isolate execution, backed by full-chain auditing and private deployment. Kingsoft also became the first AI-office vendor to receive an "agent identity code" — a verifiable digital ID for its agents.
"Models and algorithms are the brain; authoritative corpus is the knowledge reserve. AI software needs proper education to shape its values." — Wang Shaokang, VP of Kingsoft Office
Standardization accelerates: from pilot to scale
With the cyberspace administration and NDRC having issued deployment guidelines for government LLMs earlier, this technical standard marks the shift from "can we use it" to "how to use and govern it well." For industry watchers the signal is clear: government is becoming the first LLM vertical to scale — with standards first and safety as the ticket of entry. Other industries would do well to study this playbook.
(Information synthesized from China Daily, Ifeng and Zhonghong reports)