AI × Entertainment Regulation — Comparison by Country¶
⚠️ General information, not legal advice. Base practical decisions on the latest primary sources and professional review. A cross-cutting, issue-by-issue overview is in ethics-legal.md (Japanese). This page rearranges it by country and region for comparison. Japanese edition: ../../../ethics-legal/by-country/README.md
Comparison table (as of September 2026)¶
| Issue | Japan | US | EU | China | South Korea |
|---|---|---|---|---|---|
| Training | Generally permitted under Copyright Act Art. 30-4 (except where an "enjoyment" purpose coexists or rights holders' interests are unreasonably harmed) | Fair use being decided in litigation. A district court upheld training on lawfully acquired books | TDM exception (rights holders may opt out). General-purpose AI must publish a training-data summary | Interim Measures for Generative AI Services (Aug 2023) Art. 7: lawfully sourced data and models, no infringement of others' IP, consent for personal information | Rights holders' permission generally required. The government proposed a TDM exception, not enacted as of 2026 |
| Labeling AI outputs | No legal duty (guidelines only) | No federal law; some state laws | AI Act Art. 50: from Aug 2026, notice of interaction with AI, machine-readable marking of outputs, deepfake disclosure | Labeling Measures: from Sep 2025, both visible labels and metadata | AI Basic Act: from Jan 2026, advance notice of generative AI, labels and watermarks (fines deferred at least one year) |
| Voice & likeness | Right of publicity. The Ministry of Justice's Aug 2026 interpretive guideline states that voice is covered | State laws (e.g., Tennessee's ELVIS Act). The federal NO FAKES Act is pending | National laws and GDPR | Civil Code Art. 1023 protects voice by applying portrait-right rules. In 2024 the Beijing Internet Court held unauthorized use of an AI voice infringing for the first time | — |
| AI companions & minors | No legal duty | New York (notice to all users every 3 hours), California SB 243 (protections for minors; in-game bots excluded) | AI Act Art. 50(1) (notice that it is AI) | Anthropomorphic AI Interim Measures: from Jul 2026, "virtual partner/family" services banned for minors; notice after 2 hours of continuous use | — |
| In a word, for entertainment | Lenient on training, but increasingly strict on output similarity and voice | Litigation and settlements/licensing proceed in parallel | Most specific labeling duties; allowances in how artistic works are labeled | Strictest on labeling and protecting minors | New labeling duties; labels that do not hamper artistic enjoyment are allowed |
A realistic policy for shipping globally¶
- Design to the strictest region: for AI characters, China (minors, 2 hours) and US state laws (crisis response, 3 hours); for labeling, the EU, China, and Korea
- Build in both visible labels and machine-readable marks from the start: this satisfies EU, China, and Korea requirements together
- For voice, use a licensed real voice or one that does not imitate a real person: works under both Japan's guideline and US state laws (→ ai-characters P38)
- Conversation scope changes the legal classification: design NPCs limited to in-game topics separately from companions (→ ai-characters P01)
Tracking¶
- US: the Thomson Reuters v. ROSS appellate decision; NO FAKES Act
- EU: the final Code of Practice for Art. 50
- Japan: consideration of amending the Unfair Competition Prevention Act (voice protection)
- Korea: when the AI Basic Act's fine deferral ends; detailed labeling rules