Japan's 2024 Deepfake Laws: What Changed and What Gaps Remain

A Platform Law, Not a Deepfake Law

On May 17, 2024, the Diet passed the Act on Response to Circumstances Caused by the Distribution of Information via Specified Electronic Telecommunications, known by its Japanese shorthand as 情報流通プラットフォーム対処法, the Information Distribution Platform Act. It replaced the older Provider Liability Limitation Act and took effect in April 2025.⁴

The law never mentions deepfakes. It targets large platform operators directly: companies above a certain scale must publish clear takedown criteria, respond to removal requests within a defined window, and report on how complaints are handled. The reform grew out of a rights-infringement complaint volume that had stayed high for years, driven as much by ordinary defamation as by synthetic media.⁵

For a fraud victim, this matters in practice more than in name. A woman defrauded by a fake Horie video still has no criminal statute that names the deepfake itself. What she gains, at least on paper, is a platform now obligated to act on a takedown request faster than eventually.

A Guideline, Not a Law

Weeks earlier, on April 19, 2024, Japan's Ministry of Economy, Trade and Industry and Ministry of Internal Affairs and Communications released the first version of the AI Guidelines for Business (AI事業者ガイドライン), folding three earlier sets of AI principles into a single document covering developers, providers, and users of AI systems.⁶

The guidelines ask AI providers to weigh misuse risks, including impersonation and disinformation, and to design in safeguards. They carry no penalty for being ignored. That is deliberate. Japan's approach to AI governance runs on non-binding soft law that regulators can revise without a legislative fight, in contrast to the EU's binding AI Act, which passed the same month. The tradeoff: a platform or developer that decides the guidelines are not worth following faces no legal consequence for that choice.

What Still Isn't a Crime

Creating a deepfake of a real person remains, by itself, legal in Japan under current statute. What is illegal is what happens after: defaming someone under Article 230 of the Penal Code, defrauding them under the fraud statute, or interfering with their business. Horie's impersonators were prosecutable for fraud, not for cloning his voice.

It is the same structural choice that runs through Japan's handling of non-consensual intimate deepfakes, covered in more detail in our earlier piece on portrait rights: the law reaches downstream harm, not the act of generation itself.

Elections: The Law That Still Isn't There

Two years earlier, in late 2023, a synthetic clone of Prime Minister Fumio Kishida's voice reading fabricated, vulgar remarks spread across social media, drawing more than two million views before platforms took it down.⁷ As of mid-2026, Japan still has no election law provision requiring that AI-generated campaign content be labeled as such.

The Liberal Democratic Party's AI and Web3 project team proposed amendments in 2026 to both the Public Offices Election Act and the Information Distribution Platform Act that would require disclosure labels on any AI-generated or AI-altered image or video used in a campaign, including so-called disqualification campaigns aimed at damaging a candidate. The proposal covers anyone posting online, not just candidates or parties, and gives platforms explicit legal grounds to remove undisclosed synthetic content quickly. As proposed, it carries no criminal penalty.⁸

That leaves Japan well behind South Korea, which amended its own election law in 2023 to ban the production and distribution of AI-generated campaign videos outright during the 90 days before an election. Violators face up to seven years in prison or a fine of 50 million won.⁹ Japan's proposal, as of mid-2026, is still under Diet review.

What This Means for Now

2024 gave Japan two real levers: a platform takedown law and a soft-law guideline. Neither targets deepfakes by name, and the fraud wave that exposed the problem has not slowed. NPA case counts for SNS-based investment and romance fraud kept climbing into 2025.¹⁰

Until the election-disclosure law passes and a dedicated deepfake statute exists, the working defenses are the same ones that applied before 2024. Treat any investment pitch from a familiar face on social media as unverified until you confirm it through a channel that person actually controls. Assume a platform ad featuring a celebrity has not been vetted by that celebrity. And if you are defrauded by one, it is the fraud statute, not a deepfake law, that gets a case opened.