What was actually decided
On June 12, 2026, Prime Minister Sanae Takaichi chaired the 55th meeting of the Intellectual Property Strategy Headquarters and formally adopted the Intellectual Property Strategic Program 2026 (kantei.go.jp). The plan sets three priorities: concentrating IP investment across 17 designated strategic technology fields, building "principles and codes" on AI transparency and rights protection, and clearing bottlenecks in the content industry's talent, production, and distribution pipeline. Takaichi framed the AI plank in narrow terms, saying the goal is to address concerns that AI-generated content could infringe on others' rights while still allowing "safe and secure use of AI."
Two threads inside that plank matter most for the AI industry: a commitment to develop compensation frameworks for generative-AI copyright use, and continued legislative discussion of rules against AI-generated voice and likeness imitation of actors and voice performers.
The compensation track is the more advanced one
Japan's starting position is unusually permissive. The 2018 amendment to the Copyright Act created a flexible-use exception — including for machine learning and AI training — when the use doesn't unreasonably conflict with a rights holder's market (bunka.go.jp). That carve-out is one reason Japan has been described as running one of the world's most AI-training-friendly copyright regimes. The 2026 Program does not touch that exception. Instead, it directs work toward mechanisms that would compensate rights holders for AI-related harms and considers civil remedies, including collective enforcement for cases that touch large numbers of creators — the kind of harm one individual manga artist or novelist has little practical ability to litigate alone.
This is a bargain, not a rollback. Analysts at MLex describe it as Japan trying to keep AI training "largely protected" while adding pressure through disclosure requirements and market-based compensation talks rather than restrictive new statutes (MLex). The Cultural Affairs Agency has been running exactly this kind of dialogue since 2024, convening AI developers (including Google and Microsoft) and creator groups like the Japan Cartoonists Association to work out dataset licensing and revenue-sharing models before any court forces the issue.
Voice and likeness protection is earlier-stage — and legally murkier
The second track is less settled. Japanese case law recognizes a right of publicity over a person's name, image, and voice, but it derives from constitutional and civil-code precedent rather than a dedicated statute, and the Unfair Competition Prevention Act's protection for "well-known indications" has historically required something closer to a trademark-like identifier — not just a talent's likeness on its own. That gap is precisely why a wave of voice actors, including Kenjiro Tsuda, have pursued individual suits and public complaints over AI tools trained on their voices without consent. The 2026 Program commits only to continued discussion, including possible amendments to the Unfair Competition Prevention Act — not a drafted bill, let alone an enacted one.
Steelmanning the case for faster, harder rules
The strongest argument for moving quickly here is that voice and likeness harms are irreversible in a way copyright infringement often isn't. A pirated manga panel can be taken down; a synthetic voice clone that has already been used to narrate thousands of unauthorized clips has already done its reputational and commercial damage, and no compensation formula restores a performer's ability to control how their own voice is used. Voice actors, unlike novelists or illustrators, sell something inseparable from their identity — and Japan's seiyuu industry is a global export product precisely because that identity carries commercial weight. A framework that leaves this to years of stakeholder dialogue risks the same outcome the EU and California have already tried to preempt: courts filling the gap ad hoc, with inconsistent results, while the technology that would be regulated keeps shipping.
Why the go-slow approach is still the right call
Even granting that harm, Japan's sequencing is defensible. Legislating a personality-rights regime for AI outputs before the underlying case law has matured risks freezing in place definitions — of "imitation," of "consent," of what counts as a protected voice — that a live dialogue between platforms, talent agencies, and regulators can refine faster than a statute can be amended. The compensation-first approach lets Japan test whether licensing markets for AI training data can emerge voluntarily, which is cheaper to unwind than a broad new liability regime if it turns out to chill smaller AI developers who can't afford collective-licensing overhead. Given that Japan is trying to build a competitive domestic AI sector inside the same plan that protects performers, holding the line on flexible-use copyright while working out compensation and publicity-rights details through structured dialogue is a more proportionate first move than legislating first and correcting later.
The test now is speed. A "continued discussion" that stretches past the 2027 plan cycle without draft text will look, in retrospect, like deference to AI developers dressed up as deliberation.