Horizontal AI regulation

How the four jurisdictions compare

	EU	USA	Canada	China
Approach	One horizontal law, the AI Act, graded by risk	No federal AI law; executive orders, voluntary frameworks, sector regulators and state laws	No federal AI law since AIDA died in 2025; directive for the federal government, privacy law, voluntary code	Technology-specific regulations for algorithms, deep synthesis, generative AI and labeling
Binding on companies	Yes, for providers, deployers, importers and distributors	Only through state laws and existing sector law	Only through privacy and sector law	Yes, for services offered to the public in China
Risk logic	Prohibited, high-risk, transparency, minimal risk	Voluntary risk management (NIST AI RMF)	Impact levels I–IV for federal systems	Public-opinion attributes and social mobilization capacity
Documentation	Technical documentation, instructions for use, logs	Frontier AI frameworks and transparency reports (California)	Algorithmic Impact Assessment (federal institutions)	Algorithm filing and security assessment
AI-generated content	Disclosure and machine-readable marking	Some state laws; no general federal duty	Voluntary	Mandatory explicit and implicit labels
AI in machinery and products	Through product law: Annex I Section A directly, Section B (incl. machinery since 2026) via sector law; see the machinery AI timeline	Sector regulators (FDA, NHTSA) and OSHA duties; consensus standards	Sector regulators and provincial occupational safety law	GB standards and CCC certification; AI rules target online services
Main authority	AI Office and national market surveillance authorities	Sector agencies, state attorneys general	Treasury Board (public sector), privacy commissioners	Cyberspace Administration of China (CAC)

How it applies

• Technical documentation: Under a horizontal law, documentation duties are the same across industries — an HR tool and a medical device both need technical documentation and instructions for use if they are high-risk. Sectoral and vertical regimes produce different, overlapping documents: a filing in China, a transparency report in California, an impact assessment for a Canadian federal agency.
• Global products: Companies selling in all four markets usually build one documentation set to the strictest regime — in practice the EU AI Act — and add jurisdiction-specific artifacts such as the Chinese algorithm filing or content labels.
• AI and retrieval: Regulatory answers depend on jurisdiction. Tag regulatory content by jurisdiction and date; a question about “AI transparency obligations” has four different answers.

Horizontal vs. vertical regulation

A horizontal law defines obligations by risk and role, independent of sector. Even the EU's horizontal law leans on sector law for products: since the Digital Omnibus on AI, AI in machinery is assessed under the Machinery Regulation, and product-level transparency is handled by separate instruments such as the digital product passport, compared in the product transparency overview. A vertical (or sectoral) approach regulates specific technologies or uses — China regulates recommendation algorithms, deep synthesis and generative AI in separate rules; the USA relies on agencies such as the FTC, FDA or EEOC applying existing law to AI in their fields.