China’s AI rules decide launch dates. Europe’s will too
By Collin Hogue-Spears, Independent
On 15 August 2023, China’s Interim Measures for Generative AI Services took effect. Sixteen days later, Baidu opened Ernie Bot, its generative AI assistant, to the public. The speed was not improvised. Baidu had spent years building content moderation, security review, and documentation infrastructure for its search engine and earlier AI products, then repurposed that evidence engine for the newer filing and generative AI requirements. Filing, assessment, documentation, and interface obligations were in place before the general public was able to touch the service. The system was ready before the product was.
While researching my book From Lab to Life: How AI Works in China, I kept encountering the same operational pattern: capability, compliance, and distribution moved as one system, and the teams that treated governance as engineering work shipped first. No one needs to ask Europe to copy China’s rules; the AI Act has already built its own version of the mechanism, a review that sits inside the release path. The open question belongs to the teams the Act covers: can they produce the evidence when the deadlines start landing? What follows is how China made that routine, where the Act places the same checkpoints, and three moves to make before the dates hit.
The Gate Inside the Release Path
China’s generative AI regime operates through interconnected mechanisms. A service that can shape public opinion, the trigger category in the rules, must file with the regulator and pass a security assessment before launch, then display its filing information once live. The scale is past the experimental stage: 868 large-model services had completed filing as of April 2026, per the Cyberspace Administration of China (CAC)’s June 2026 update. Behind that number are hundreds of product teams running the same pre-launch process. Filing is now a standard step in shipping an AI product in China.
The most concrete build gate is labeling. The problem it solves is recognition: a user cannot tell synthetic content from human-made content, so the rules make the content announce itself. China’s synthetic content labeling measures, effective 1 September 2025, require explicit on-screen labels, implicit metadata labels, and checks by app distribution platforms when AI apps are listed, per the CAC labeling measures. This is where governance stops being a policy doc and becomes sprint work: where the label renders, how metadata survives an export, and which team owns the check.
Enforcement closes the loop. By 6 July 2026, the first stage of China’s 2026 Clear and Bright campaign had handled more than 14,000 non-compliant AI products and services, per Xinhua’s report. Once a regulator can act at the point of distribution, missing documentation becomes a launch risk.
Europe’s Version of the Turnstile
Neither regime is one-size-fits-all. China scales obligations by reach; the AI Act scales by risk tier. In both, classification decides which credentials a product owes at release.
Europe holds the same lever through different machinery. Under Article 83 of the AI Act, a market surveillance authority that finds a missing declaration of conformity, a missing database registration, or unavailable technical documentation orders the provider to remedy the issue; where the failure persists, it restricts, prohibits, withdraws, or recalls the system. The cadence differs from China’s campaign sweeps; the endpoint is written into the statute.
The calendar is already live. Prohibited practices took effect in February 2025, and general-purpose AI (GPAI) obligations took effect in August 2025, per the European Commission. Transparency rules under Article 50 arrive in August 2026, with a transition to 2 December 2026 only for systems already on the market, which puts them next on the build calendar for any system that interacts with people or generates content that must be identifiable. The high-risk dates moved: under the Digital Omnibus, the EU’s simplification package, obligations shift to 2 December 2027 for stand-alone high-risk systems and 2 August 2028 for AI embedded in regulated products. The European Parliament approved the package in June 2026, and the Council gave final adoption on 29 June, locking the new calendar in place.
The standards gap is the bridge to watch. CEN and CENELEC’s JTC 21, the committee drafting the Act’s harmonized standards, is working across ten areas from risk management to conformity assessment, and the first candidate standard entered public inquiry on 30 October 2025, per the Commission’s standardization page. Once cited in the Official Journal, those standards create a presumption of conformity. They will do for the AI Act what filing templates and labeling standards did in China: translate legal text into engineering instructions.
The labeling measures also show how China handles overlapping authority: four agencies spanning content, telecoms, security, and media issued a single framework, so product teams manage a single combined obligation rather than four separate policies. European teams face the same coordination problem across the AI Act, data protection, and product safety law.
Three Moves for the Next Twelve Months
First, map the gates onto the product lifecycle. Mark which launch step depends on classification, conformity assessment, GPAI documentation, transparency notices, labeling, or database registration. A gate that stays invisible in the lifecycle reappears later as a delay.
Second, build a standards-delta file. Teams do not need to wait for the Official Journal. List current controls for the ten JTC 21 areas, assign an owner, an evidence artifact, and a release impact to each, and review the file as the standards mature. That turns a legal unknown into an engineering queue.
Third, make labeling and documentation product features. Article 50 lands in August 2026. China’s experience since September 2025 shows that labels, metadata, and platform checks hold up only when product, engineering, legal, and risk teams build them together. Automate the evidence capture wherever possible: a pipeline built for the AI Act can be pointed at the next regime, just as Baidu’s older compliance infrastructure absorbed the generative rules in sixteen days. In regulated AI markets, compliance velocity has started to determine product velocity.
Where to go Deeper
I will present “How China Made AI Governance Operational: Lessons for European Practitioners” at IRM UK’s Data & AI Conference Europe 2026 in London on 4 November, a 20-minute session on which of China’s governance patterns transfer to enterprise AI programs without importing China’s regulatory model. The session draws on regulatory sources and research for my forthcoming book, From Lab to Life: How AI Works in China.
About the author: Collin Hogue-Spears is an independent researcher and author of From Lab to Life: How AI Works in China (Gatekeeper Press, 4 August 2026), drawing on 20 years of experience across technology, product, and compliance across the US, Europe, and China. His commentary has been quoted in The Wall Street Journal, Politico Pro, and Dark Reading; find out more at collinhoguespears.ai.
Join Collin Hogue-Spears at Data & AI Conference Europe 2026
As organisations prepare for the EU AI Act and an increasingly complex regulatory landscape, understanding how AI governance can become part of everyday operations is more important than ever. Collin Hogue-Spears’ session offers a unique perspective on how governance can move beyond policy documents and become an integral part of building, deploying and managing AI systems.
Join Collin Hogue-Spears on Wednesday 4 November for “How China Made AI Governance Operational: Lessons for European Practitioners,” part of the Governance, Risk, Privacy & Responsible AI track at Data & AI Conference Europe 2026.
Whether you’re responsible for AI governance, compliance, data governance or AI strategy, you’ll leave with practical insights into operational governance patterns that can help organisations navigate regulatory change with greater confidence.
Explore the full agenda, discover more expert speakers and book your place at Data & AI Conference Europe 2026 today.
