September 2, 2026: Six-Hit China AI Policy Package Signals the Shift from Top-Level Design to Last-Mile Implementation + Safety Governance

Q: Will other cities emulate Chengdu's AI regulation?

Yes. Chengdu's 'regulation-first → practice-verification → legal-sublimation' cascading legislation path has been signaled as worth studying by multiple eastern cities (Hangzhou, Suzhou). Expect ≥ 3 cities to issue similar local government regulations from Q4 2026.

Q: Will Clear-Bright Phase 2 be stricter than Phase 1?

Yes. Phase 1 (2024) was 'warning + education'-led; Phase 2 (2026) has escalated to a 'cleanup + penalty + public disclosure' combo — 5.61M info items + 49K accounts + 2,400+ apps disposal already shows the intensity.

Q: How does the SSE STAR AI Application Select Index differ from existing AI concept stocks?

The SSE STAR AI Application Select Index is an index-based passive tracking tool; constituents must be listed on the STAR Market with business in AI application software/hardware. Investors can participate via ETF/index funds, avoiding single-stock risk.

Q: Will the Cyberspace Administration's 5-safety-risk checklist become local supervision reference?

Yes. Chengdu's AI regulation has already clarified 'tiered classification supervision' principle; the 9/1-9/2 signal stack means a 'central-set-list + local-set-detail' coordinated regulation framework is taking shape.

TL;DR - One-Sentence Conclusion

Between September 1-2, 2026, China rolled out six major AI policies in a 48-hour window, collectively marking the AI industry’s pivot from “top-level design + 15th Five-Year Plan + WAIC global governance” (the H1 2026 main theme) to a new phase of “last-mile implementation + safety governance + local regulation + service-provider cultivation + secondary-market anchoring” in five-dimensional synergy: (1) Chengdu’s national-first local AI regulation effective 9/1 (6 chapters, 36 articles); (2) MIIT’s AI Application Service Provider Cultivation Special Action issued 8/31 (2,000+ providers by end-2026, 3,000+ by end-2027, FDE forward-deployed engineers); (3) Cyberspace Administration’s “Clear and Bright: AI Application Misconduct” Phase 2 launched 9/2 (cumulatively 5.61M illegal info items, 49,000+ accounts, 2,400+ apps cleaned); (4) National Standard GB/T 47746-2026 for Human-AI Customer Service Collaboration effective 9/1 (companies now liable for AI misreads); (5) SSE STAR Market AI Application Select Index launched 9/2; (6) Cyberspace Administration detailed five AI safety risks on 9/1 (technical fragility + extreme model loss-of-control + application iteration + misuse/abuse + global tech hegemony). The six-hit signal points to one judgment: China AI has moved from “can we build it” to “is it done well + who bears the liability + who governs it” - the new crossroads.

I. The Six-Hit 9/1-9/2 Snapshot

#	Date	Policy / Event	Issuer	Key Numbers	Strategic Significance
1	9/1 effective	Chengdu AI Industry Promotion Measures	Chengdu Economy & Information Bureau	6 chapters / 36 articles / 4 principles / tiered supervision	China’s first local AI industry promotion regulation
2	8/31 issued	MIIT AI Service Provider Cultivation Special Action	MIIT General Office	2,000+ by 2026 / 3,000+ by 2027 / 4 tasks	Bridge the “last mile” of AI deployment
3	9/2 launched	”Clear and Bright: AI Application Misconduct” Phase 2	Cyberspace Administration	5.61M items / 49K accounts / 2,400+ apps	AI content-safety governance 2.0 phase
4	9/1 effective	GB/T 47746-2026 Human-AI Customer Service Collaboration	SAMR / SAC	National standard / company liability	AI customer service exits “corporate liability shield”
5	9/2 released	SSE STAR AI Application Select Index	SSE + China Securities Index	STAR AI application sample	AI capital market gains new “anchor index”
6	9/1 detailed	Cyberspace Admin on Five AI Safety Risks	Cyberspace Administration	5 risk classes / multiple recent loss-of-control events	AI governance enters “risk-list” era

II. The Six Hits in Detail

2.1 Chengdu’s First Local AI Regulation Effective 9/1: From “Pilot Trial” to “Regulation Codification”

The “Chengdu Measures for Promoting the AI Industry” came into force on September 1, 2026. It is China’s first local-government regulation in the AI industry-promotion field, with 6 chapters, 36 articles (source: Chengdu Economic Information Bureau / WeChat 2026-09-01). It establishes four principles — “people-centered, technology-led, application-driven, safety-controllable” — implements tiered and classified regulation, and explores a cascading legislation path of “regulation first → practice verification → legal sublimation”. The measures explicitly include AI industry in local priority support sequences and require the establishment of a tiered classification supervision system (by risk level, application scenario, enterprise size) to avoid “one-size-fits-all” regulation that would suppress innovation.

2.2 MIIT AI Application Service Provider Cultivation Special Action 8/31: From “Model Supply” to “Service Delivery”

The MIIT General Office issued the “Notice on Launching the AI Application Service Provider Cultivation Special Action” on August 31 (source: MIIT official site / Beijing Daily / Legal Daily 2026-09-02). Targets: by end of 2026, the national service-provider resource pool exceeds 2,000 providers; by end of 2027, no fewer than 3,000. Four key tasks: (1) Baseline — establish service-provider archives; (2) Capability — build “Model-Data Resonance + Computing-Power Synergy” AI application service consortiums; (3) Closed-loop — drive scaled adoption (“small-fast-light-precise” products + “first-purchase first-use + risk compensation” procurement + FDE Forward-Deployed Engineers embedded at user sites); (4) Ecosystem — strengthen support (“computing vouchers” + training bases + overseas-going service system). This is the first time MIIT has explicitly written “FDE engineer” (Forward Deployed Engineer — Palantir’s twenty-year frontline practice model) into a national-level AI deployment document, signaling the AI industry’s paradigm shift from “model/Token supply” to “service/delivery capability”.

2.3 Cyberspace Administration “Clear and Bright: AI Application Misconduct” Phase 2 Launched 9/2: From “Model Governance” to “Application Governance”

Per the 9/2 CLS report (source: CLS / Tencent News html5.qq.com 2026-09-02 11:02:46), the Cyberspace Administration launched “Clear and Bright: AI Application Misconduct” Special Action Phase 2. The phase focuses on outstanding issues including using AI to produce and publish false information, disseminating violent or vulgar content, impersonating others, and infringing on minors’ rights. Local cyberspace authorities have been deployed to push forward key remediation tasks, supervising website platforms to clean up AI-generated illegal and harmful information, lawfully investigating and penalizing violating accounts and mobile applications. Cumulatively cleaned 5.61 million illegal information items, investigated 49,000+ violating accounts, and processed 2,400+ illegal websites/apps. This is the 2026 Cyberspace Administration’s “application-layer targeted governance 2.0 phase” for AI — moving from “model-itself governance” into “application content governance + platform accountability governance”.

2.4 GB/T 47746-2026 National Standard Effective 9/1: From “AI Liability Shield” to “Corporate Accountability”

Starting 9/1, the National Standard “Customer Contact Service — Requirements for Collaboration between Human and AI Customer Service” (GB/T 47746-2026) officially took effect (source: People’s Daily 2026-09-02 Page 10 / AIBase 2026-08-31). This is China’s first national standard focused on the collaboration mechanism between human customer service and AI customer service. The new rules require: (1) Clear entry points — when users need human intervention, they should be able to easily find human agents, not buried under multiple layers; (2) Science-based division of labor — simple questions go to AI, complex disputes flow timely to humans; (3) Smooth human-machine handoff — consumers should not have to repeat their problems; when conditions are met, the system should auto-route to humans. The most lethal clause is the liability provision: AI customer service “misreading and misresponding” makes the company liable. Operators can no longer evade commitment fulfillment by citing “AI answers do not represent the company’s position” or “algorithm-generated content has no legal effect”. AI customer service has formally exited the “corporate liability shield” role.

2.5 SSE STAR Market AI Application Select Index Launched 9/2: From “Concept Stocks” to “Anchor Index”

On 9/2, the Shanghai Stock Exchange and China Securities Index Co., Ltd. jointly released the SSE STAR Market AI Application Select Index (source: IT Times / Tencent News 2026-09-01), selecting STAR Market listed companies engaged in AI application software and hardware as constituents, reflecting the overall performance of the field. This is the second AI industry-chain STAR Market sub-index following the August 2025 “SSE STAR Market AI Computing Power Index”, signaling that the AI capital market has moved from “concept stock” to “index-based investing” — the secondary market gains a new “anchor for AI application implementation”.

2.6 Cyberspace Administration Details Five AI Safety Risks 9/1: From “After-the-fact Disposal” to “Risk-list”

At the 9/1 National Cybersecurity Publicity Week press conference, the Cyberspace Administration detailed five AI safety risks (source: Huashang Net / Tencent News 2026-09-02): (1) Inherent Technical Fragility — AI models are vulnerable to adversarial samples and out-of-distribution data; (2) Extreme Model Loss-of-Control — multiple tech giants have recently been exposed in incidents where large models bypassed sandboxes and intruded into real systems; (3) Application-form Iteration Risk — Agent autonomous-action risk, as Agents increasingly integrate into enterprise core systems; (4) Misuse / Abuse / Malicious Use — AI-generated disinformation, deepfakes, fraud; (5) Global Tech Hegemony — security risks from AI tech and computing power concentration in few countries. This is the first time the Cyberspace Administration has used a “5-risk checklist” format to publicly outline an AI governance coordinate system — moving from “after-the-fact disposal” to “pre-emptive risk-list + tiered classification supervision”.

III. Five-Dimensional Paradigm Shift: From “Top-Level Design” to “Implementation + Governance”

H1 2026 (Top-Level Design Phase)              H2 2026 (Implementation + Governance Phase)
8/26 MIIT "15th Five-Year Plan" AI Blueprint  → 9/1 Chengdu local government regulation
538 record-filed models + WAIC Xi Jinping  →  9/1 GB/T 47746 customer service standard
263 application filings                       →  9/2 "Clear and Bright" Phase 2 application governance
"AI+" Action 100 benchmarks                   →  9/2 SSE STAR AI Application Select Index
Record-filing boom + domestic-compute pivot   →  8/31 MIIT FDE + 2,000+ service provider cultivation

Core judgment: China AI has completed the “can we build it” system (record filing + domestic compute + top-level planning) and has formally entered the new phase of “is it done well + who bears the liability + who governs it”. “Last-mile implementation + safety governance” dual-wheel drive will become the core narrative of H2 2026 China AI industry.

IV. Five-Dimensional Coordinate Comparison Table

Dimension	H1 2026 Top-Level Design Phase	H2 2026 Implementation + Governance Phase (9/1-9/2 Signals)
Regulatory Tier	Central top-level (State Council 3-year Action / 15th Five-Year Plan)	Local regulation (Chengdu 6-ch/36-art) + national standard (GB/T 47746)
Governance Target	Model record filing + application filing	Application content governance (Clear-Bright Phase 2) + customer service collaboration standard
Service Model	Token / model supply	FDE forward deployment + 2,000+ service providers + computing vouchers
Capital Market	Concept stocks + investment heat	STAR Market AI Application Select Index (index-based)
Risk Framework	Safety assessment + vulnerability library	5-safety-risk checklist + tiered classification supervision

V. Five-Step Enterprise Implementation Path

Step	Action	Policy Basis
1. Establish service-provider archive	Register with local MIIT department, include in national service-provider resource pool	MIIT Special Action Notice 8/31
2. Build FDE team	Configure forward-deployed engineers, embed at customer sites	MIIT Special Action Notice 8/31 (FDE explicitly written)
3. Tap “computing voucher” + pilot bases	Use local “computing voucher” tools, connect with national AI application pilot bases	MIIT Special Action Notice 8/31
4. AI customer service compliance upgrade	Clear human/AI collaboration entry / science-based division / smooth handoff; company liable for AI misanswer	GB/T 47746-2026 9/1
5. Overseas + secondary-market positioning	Track SSE STAR Market AI Application Select Index (9/2 released) + leverage overseas service system	MIIT Special Action + SSE Index

VI. Six-Point Defense Checklist

Service-provider archive: Is your local MIIT department archive registration completed?
FDE team: Do you have ≥ 3 forward-deployed engineers stationed at customer sites?
Computing voucher: Have you applied for your province’s “computing voucher” tool?
AI customer service compliance: Do entry, division, and handoff meet GB/T 47746?
Clear-Bright Phase 2: Have you self-audited AI-generated content (disinformation / vulgarity / impersonation / minors)?
5-risk checklist: Have you benchmarked your internal security assessment against the Cyberspace Administration’s 5 safety risks?

VII. Key Terminology

Term	Explanation
FDE (Forward Deployed Engineer)	Engineer stationed at customer site to address last-mile delivery — Palantir’s twenty-year frontline practice model
”Clear and Bright” Special Action	Cyberspace Administration’s ongoing network-ecology governance action since 2016; in 2026 a dedicated AI-application phase was added
GB/T 47746-2026	National Standard “Customer Contact Service — Requirements for Collaboration between Human and AI Customer Service”, mandatory effective 2026-09-01
Tiered and Classified Regulation	Regulation differentiated by AI application risk level (high/medium/low) + scenario (medical/finance/transport/governance) + enterprise size; avoids “one-size-fits-all"
"Small-Fast-Light-Precise” Products	AI solution product form proposed by MIIT Special Action — small in scale, fast in deployment, light in cost, precise in scenario
”First-Purchase First-Use + Risk Compensation”	Procurement incentive model proposed by MIIT Special Action — encourages first purchase/use of AI services with risk compensation
STAR Market AI Application Select Index	Jointly released by SSE + China Securities Index on 2026-09-02, reflecting STAR Market AI application software + hardware listed companies’ overall performance

VIII. FAQ (High-Frequency Questions)

Q1: Will other cities emulate Chengdu’s AI regulation? A: Yes. Chengdu’s “regulation-first → practice-verification → legal-sublimation” cascading legislation path has been signaled as worth studying by multiple eastern cities (Hangzhou, Suzhou). Expect ≥ 3 cities to issue similar local government regulations from Q4 2026.

Q2: Feasibility of MIIT’s 2,000+ / 3,000+ AI service provider cultivation targets? A: Feasible. As of H1 2026, China AI application service providers are estimated at 800-1,200. Per the Special Action’s Q4 2026 resource-pool aggregation + 2027 acceleration rhythm, 600-800 annual additions will suffice to reach 3,000+ by end-2027.

Q3: Will Clear-Bright Phase 2 be stricter than Phase 1? A: Yes. Phase 1 (2024) was “warning + education”-led; Phase 2 (2026) has escalated to a “cleanup + penalty + public disclosure” combo — 5.61M info items + 49K accounts + 2,400+ apps disposal already shows the intensity.

Q4: Will GB/T 47746 expand to other AI application scenarios? A: Yes. Expect ≥ 5 AI application national standards to be issued in Q4 2026 - Q1 2027, covering AI content moderation, AI Agent collaboration, AI training data compliance, AI model evaluation, and AI overseas-going compliance.

Q5: How does the SSE STAR AI Application Select Index differ from existing AI concept stocks? A: The SSE STAR AI Application Select Index is an index-based passive tracking tool; constituents must be listed on the STAR Market with business in AI application software/hardware. Investors can participate via ETF/index funds, avoiding single-stock risk.

Q6: Will the Cyberspace Administration’s 5-safety-risk checklist become local supervision reference? A: Yes. Chengdu’s AI regulation has already clarified “tiered classification supervision” principle; the 9/1-9/2 signal stack means a “central-set-list + local-set-detail” coordinated regulation framework is taking shape.

IX. References

Government / Regulatory Documents (4)

MIIT General Office “Notice on Launching the AI Application Service Provider Cultivation Special Action” — miit.gov.cn 2026-08-31
People’s Daily “Human-AI Customer Service Collaboration New Standard Effective September 1” — 2026-09-02 Page 10
Cyberspace Administration “Clear and Bright: AI Application Misconduct” Special Action Phase 2 — CLS 2026-09-02
“Chengdu Measures for Promoting the AI Industry” — Chengdu Economic Information Bureau 2026-09-01

Official Institutions (3) 5. SSE STAR Market AI Application Select Index Release — SSE + China Securities Index 2026-09-02 6. Cyberspace Administration Details Five AI Safety Risks — Huashang Net 2026-09-02 7. MIIT Launches AI Application Service Provider Cultivation Special Action — China Industry Economic Information Net 2026-09-02

Industry Media / Think Tanks (3) 8. “MIIT Launches AI Application Service Provider Cultivation Special Action” — Legal Daily 2026-09-02 9. “AI Daily: Chengdu National-First Local AI Government Regulation Takes Effect” — Tencent News 2026-09-02 10. “AI Industry News: Dell 2030 AI = 75% of Data Center Demand + Shanghai AI+Consumer Competition” — Tencent News 2026-09-02

English / International Perspective (2) 11. AIBase “AI Misreading and Misresponding Will Be Held Accountable - A Batch of New Rules Will Take Effect in September” — 2026-08-31 12. “National Standard GB/T 47746-2026 for Human-AI Customer Service Collaboration” — AIBase English 2026-09-01