Japan’s AI Promotion Act took effect May 28, 2025. What made it notable then was what it lacked: no financial penalties for non-compliance, enforcement through soft-law guidance rather than regulatory sanctions. That framework was covered in full when the act passed; this brief covers what nearly a year of implementation looks like.
The short answer: a guidelines update is reportedly coming, and legal pressure on the copyright exception is building.
According to industry reporting from mailmate.jp, Japan’s Ministry of Economy, Trade and Industry is preparing a version 1.2 update to its AI Guidelines for Business, reportedly addressing cross-border data flows and interoperability with the EU AI Act. METI has not published an official announcement confirming this update or its scope. This is reported, not confirmed from a primary government source, and should be treated accordingly. METI’s guidelines carry significant practical weight in Japan even without penalty-backed enforcement, companies operating in Japan use them as the de facto compliance benchmark for the AI Promotion Act.
If the reported update addresses EU AI Act interoperability, that’s a substantive development. Japan-based or Japan-operating AI companies that also have EU exposure may find that a METI guidelines update creates a clearer path to dual compliance, or reveals where the two regimes diverge.
The copyright exception under pressure. Japan’s Copyright Act Article 30-4 is the provision that made Japan attractive to AI developers training on large datasets. It permits AI training on copyrighted material, subject to the “unreasonably prejudice” proviso, a limitation on the exception that applies when use unreasonably harms the copyright holder’s interests. As analysis from law.asia identifies, the “unreasonably prejudice” proviso is now the provision legal experts flag as the most likely litigation target as AI training use cases at commercial scale expand.
This represents a meaningful shift in Japan’s copyright risk profile. The training data permission isn’t disappearing, Article 30-4 remains in force. But its practical boundaries are untested in court at the scale of modern foundation model training. The proviso’s meaning will be shaped by litigation when it arrives, and legal experts are signaling that arrival is more a matter of when than if.
Why it matters. For AI companies that chose Japan as a training data jurisdiction partly because of Article 30-4’s permissive framing, the litigation signal is worth tracking. A court ruling that interprets the “unreasonably prejudice” proviso broadly could narrow the exception significantly without any legislative change to the act itself. The soft-law enforcement model doesn’t insulate companies from civil copyright litigation by rights holders, those are separate legal tracks.
The copyright proviso litigation signal also connects to a broader pattern. As the “Two Copyright Cases, Two Jurisdictions” brief documented, AI copyright exposure is being defined through litigation across multiple jurisdictions simultaneously. Japan is now part of that pattern, not an exception to it.
What to watch. Two near-term signals: (1) an official METI publication confirming or clarifying the version 1.2 guidelines update, if confirmed, review the scope of the cross-border data flow provisions; (2) any filed litigation specifically invoking the “unreasonably prejudice” proviso against an AI company, that would be the test case the legal community is anticipating. The approaching one-year anniversary of the act’s enactment in late May 2026 may prompt official government commentary on implementation progress.
TJS synthesis. Japan’s soft-law AI framework was designed to give the market room to develop before regulatory lines hardened. That room still exists, but the copyright exception that made Japan useful for AI training is now attracting the litigation scrutiny that soft law can’t prevent. Companies relying on Article 30-4 should be monitoring Japanese court filings, not just METI guidance updates.