MIIT sets 2028 targets for AI in software

China's Ministry of Industry and Information Technology (MIIT) issued the notice on the "AI plus software" special action implementation plan. The Chinese title is 《"人工智能+软件"专项行动实施方案》 (translation ours). The notice, numbered 工信部信发〔2026〕209号, is dated 2 September 2026. MIIT posted it at 17:24 Beijing time (UTC+8) on 11 September 2026.

The notice goes to provincial industry and IT departments and asks them to implement the plan. The full text is in the attached plan PDF.

Targets for 2028 and 2030

The plan sets these targets for 2028 (translation ours):

  • Rollout of AI coding tools covering 20,000 software firms above designated size.
  • 100 smart retrofit projects at software firms.
  • 100 benchmark agent software applications in key industries.
  • At least five high-quality open-source projects incubated. The press conference transcript calls them key (重点) projects.
  • At least 15 adaptation centers pairing base software with chips and servers.

By 2030, the plan says key software should be fully upgraded with AI. At MIIT's 11 September press conference, deputy director general Wang Weiwei said 2025 industry revenue reached 15.48 trillion yuan. He said daily token calls in China passed 500 trillion as of June 2026.

Security review for AI-generated code

Section 19 of the plan covers security. It calls for the following (translation ours):

  • Security review of AI-generated code, and defenses against malicious instruction injection into AI coding tools.
  • Tighter access control for development tools and code repositories.
  • Research on agent identity, trusted interconnection, data security and behavior control.
  • Agent security management rules covering development, deployment and use.
  • Intellectual property compliance for AI-generated code.

MIIT's Q&A on the plan adds that new software in key industries should get security testing before launch.

The plan also asks for an app store and a skills library for agent software, with listing review rules. It asks for standards for model as a service (模型即服务) and agent as a service (智能体即服务), translation ours. It calls for a graded maturity standard for AI coding ability.

Targets without budgets

The plan says it will use special fiscal funds but gives no amount. It encourages local governments to use existing funds and policy channels. The Q&A encourages localities to use computing vouchers (算力券, translation ours). It says they could cover a share of spending on domestic coding tools and domestic model coding services.

The Q&A says the plan aims to prevent a sudden wave of mass layoffs among software workers. The plan also asks for studies of AI's effect on software jobs.

Most measures use the verbs support (支持) or encourage (鼓励). The plan sets no penalties.

Foreign coding tool vendors and Chinese software firms

The plan tells local programs to favor software firms with measured gains in code quality and development output. Vendors selling AI coding tools in China should note the Q&A's call for safe and reliable (安全可靠) coding tools and its subsidies for domestic ones.

Teams building agents for Chinese customers should track the coming agent security rules and agent interface standards. China's earlier agent risk guidance is in China's AI Safety Governance Framework 3.0 adds a 33-risk agent annex. A separate draft on AI companions is covered in China's CAC drafts State Council rules to ban virtual partner services for minors.