// wiki · review · September 2026
AI regulation abroad
Draft article. The state of the norms is as of September 2026; the stages of application of foreign laws should be checked against official texts before a decision.
A draft for discussion. Foreign regimes are described as a map for exporting companies, not as a legal analysis of foreign law.
The short answer
Three models have taken shape in the world. The EU adopted a horizontal risk-tiered statute — with strict duties and fines. China regulates vertically: mandatory rules for generative services and content labelling. The USA gets by with state laws and sectoral rules without a federal standard. The Russian approach is support with soft requirements (the RF regulation map). For a company with a foreign audience the practical minimum is labelling: in the EU and China it is mandatory.
The EU: the AI Act
The Regulation entered into force on 1 August 2024 and applies in stages: first the prohibitions on unacceptable practices and the duties of general-model providers; from 2 August 2026 the Regulation applies in full, including Article 50 on transparency: content generated or altered by AI is labelled, deepfakes are disclosed, machine-readable marks are embedded into synthetic files. The European Commission has published a code of practice on transparency. For a Russian product with EU users, Article 50 is already a mandatory requirement.
China
Two layers: the rules for servicing generative AI (2023) — duties of service providers, and the measures for labelling synthetic content adopted in March 2025 and effective from 1 September 2025 together with the national standard GB 45438-2025. The requirements are stricter than the European ones: the label is double — visible to the user and hidden in metadata or watermarks — and mandatory for texts, images, audio and video; both generators and distribution platforms answer — including content created with foreign tools.
The USA
There is no federal AI law. Regulation is assembled by states: by spring 2026 more than one and a half thousand AI bills were under consideration in 45 states, and the wave moved from notifications to the “second generation” — algorithm audits, reporting, anti-discrimination requirements. Comprehensive laws — Colorado (CAIA), California, Illinois, New York with a local hiring law; the federal AI Labeling Act has not been adopted. Practically this means: requirements depend on the state of operation and the industry, without unified labelling.
Comparison and conclusions
| Jurisdiction | Approach | AI content labelling |
|---|---|---|
| The EU | A horizontal risk-tiered regulation | Mandatory (Art. 50, from 02.08.2026) |
| China | Sectoral measures + a standard | Mandatory, double label (from 01.09.2025) |
| The USA | States and industries, without a federal standard | Fragmentary (state laws) |
| Russia | The law on supporting AI | Voluntary; the platforms' duty is to provide a tool |
- Content export: services with a European or Chinese audience set up both labels — the technical implementation is one (embedded marks), the regimes differ.
- Supply chains: American and European clients carry their requirements over to suppliers — a Russian vendor will be asked about compliance with their regime.
- Dynamics: many initiatives are in the works (from the US federal labelling law to new acts of the EU Commission) — the map should be re-checked as of the date of the decision.
Checking the requirements for a specific product in target markets is the task of product rights and risks practice.