Japan Proposes Landmark AI Copyright Law for Creators

Japan's government has proposed groundbreaking copyright legislation that would require AI companies to obtain permission before using copyrighted works for model training, marking one of the most creator-friendly regulatory moves in the global AI landscape. The proposed law, announced by Japan's Agency for Cultural Affairs in mid-2025, represents a dramatic reversal from the country's previously permissive stance on AI training data and could influence similar legislation across Asia, Europe, and North America.

The legislation arrives at a critical moment when creators worldwide are battling tech giants like OpenAI, Google, Meta, and Stability AI over the unauthorized use of their works in training datasets worth billions of dollars.

Key Facts at a Glance

• Japan reverses its 2018 policy that broadly allowed copyrighted material for AI training without permission
• The new law would require explicit opt-in consent from creators before their works can be used in training datasets
• AI companies could face fines up to ¥100 million ($670,000) for violations
• The legislation includes a mandatory compensation framework for creators whose works are used in AI training
• Japan's move follows the EU AI Act but goes further on copyright protections specifically
• The law is expected to be formally introduced to Japan's Diet (parliament) by late 2025

Japan Reverses Course on AI-Friendly Copyright Stance

Japan's 2018 copyright amendment was once celebrated by AI developers worldwide. It broadly permitted the use of copyrighted material for computational analysis, including machine learning, without requiring permission from rights holders. This made Japan one of the most attractive jurisdictions for AI development.

That era appears to be ending. The new proposal explicitly narrows the scope of the 2018 exception, stating that large-scale AI model training constitutes a commercial use that directly impacts creators' livelihoods. The Agency for Cultural Affairs cited mounting evidence that generative AI outputs are displacing original creative works in the marketplace.

The reversal comes after intense lobbying from Japan's influential creative industries. Manga artists, anime studios, musicians, and photographers have argued that their works form the backbone of training datasets used by companies like OpenAI, Midjourney, and Stability AI — yet they receive zero compensation.

What the Proposed Legislation Actually Requires

The draft legislation introduces several key mechanisms that would fundamentally change how AI companies interact with copyrighted content in Japan. Unlike the EU AI Act's transparency-focused approach, Japan's proposal creates direct financial obligations.

The core requirements include:

• Opt-in licensing: AI companies must secure explicit permission before including copyrighted works in training datasets
• Compensation standards: A government-appointed board would establish fair market rates for different content categories
• Training data disclosure: Companies must maintain and publish detailed records of all copyrighted works used in training
• Retroactive compliance: Companies with existing models trained on Japanese copyrighted works must negotiate licensing agreements within 24 months
• Right of removal: Creators can demand their works be removed from training datasets, triggering model retraining obligations
• Output similarity protections: AI-generated content that is 'substantially similar' to specific copyrighted works would constitute infringement

The retroactive compliance provision is particularly significant. It means companies like OpenAI and Google, whose models were already trained on vast datasets including Japanese content, would need to either negotiate licenses or demonstrate their datasets excluded Japanese copyrighted works.

Global Implications Could Reshape AI Industry Economics

The financial impact on major AI companies could be substantial. Japan is home to the world's 3rd-largest entertainment and media market, valued at approximately $180 billion. Its manga industry alone generates over $5 billion annually, and Japanese anime commands a global market worth $28 billion.

If AI companies are forced to pay licensing fees for Japanese creative works, it sets a powerful precedent. Other nations with significant creative industries — the United States, South Korea, France, and the United Kingdom — may follow suit with similar legislation.

Compared to the EU AI Act, which primarily requires transparency about training data sources, Japan's proposal goes considerably further by mandating compensation. The EU approach tells companies to disclose what they used; Japan's approach tells them to pay for it.

Industry analysts estimate that comprehensive licensing obligations could add $500 million to $2 billion annually to the operating costs of major AI companies. This could accelerate the shift toward synthetic training data and proprietary content partnerships, trends already visible in deals like OpenAI's $250 million agreement with News Corp and Google's licensing arrangements with Reddit.

Creator Communities Rally Behind the Proposal

Japan's creative professionals have responded to the proposal with overwhelming support. The Japan Cartoonists Association, representing over 500 manga artists, issued a statement calling the legislation 'a necessary correction that recognizes creators as the foundation of AI innovation, not its raw material.'

The backlash against AI training practices has been particularly fierce in Japan's art communities. In 2024, over 100,000 Japanese artists signed a petition demanding stricter AI copyright protections. Several prominent manga artists, including creators whose works have been widely replicated by AI image generators, have become vocal advocates for regulation.

The proposed compensation framework addresses a core grievance. Under the current system, an artist whose entire portfolio was scraped and used to train a commercial AI model worth billions receives nothing. The new law would create a structured pathway for payment.

Tech Industry Pushback and Concerns

Not everyone welcomes the proposal. Japan's tech sector, which has been aggressively pursuing AI development to counter economic stagnation, warns that overly restrictive copyright rules could hamper innovation and push AI companies to other jurisdictions.

The Japan Association of New Economy, backed by companies like Rakuten and SoftBank, has raised several concerns:

• Retroactive compliance requirements could be technically impossible for some models
• Licensing costs may make smaller AI startups unviable
• Defining 'substantially similar' output remains legally ambiguous
• Japan could lose its competitive advantage in AI development to countries like China, which maintains looser copyright enforcement

SoftBank's CEO Masayoshi Son, who has positioned the company as a major AI investor with commitments exceeding $100 billion globally, has reportedly urged the government to consider a more balanced approach. His concern is that strict copyright enforcement could undermine Japan's $15 billion national AI strategy announced in early 2025.

However, proponents argue that a clear legal framework actually benefits AI companies by reducing litigation risk. In the United States, OpenAI, Meta, and Stability AI currently face dozens of copyright lawsuits with uncertain outcomes. A defined licensing system, while costly, provides legal certainty.

How This Compares to Other Countries' Approaches

Japan's proposal sits at the most protective end of the global regulatory spectrum. Here is how it compares to other major jurisdictions:

The United States has no federal AI copyright legislation. Courts are currently adjudicating key cases — including The New York Times v. OpenAI and Getty Images v. Stability AI — that will shape American copyright doctrine for years. The outcome remains highly uncertain.

The European Union's AI Act requires transparency about training data but does not mandate compensation. Rights holders can opt out of having their works used for training, but enforcement mechanisms remain underdeveloped.

China has taken a mixed approach, introducing some AI content regulations while maintaining relatively permissive training data policies that benefit domestic AI champions like Baidu, Alibaba, and ByteDance.

South Korea is reportedly studying Japan's proposal closely, given its own massive creative industries including K-pop and webtoons, which face similar AI training concerns.

Looking Ahead: Timeline and Next Steps

The proposed legislation faces several hurdles before becoming law. The Agency for Cultural Affairs plans to conduct a 60-day public comment period starting in late summer 2025, followed by revisions and formal submission to the Diet.

If the legislative process proceeds on schedule, the law could take effect by mid-2026, with the retroactive compliance window extending to mid-2028. This gives AI companies approximately 3 years to adapt their practices.

For global AI developers, the message is clear: the era of freely scraping the internet for training data is ending. Whether through legislation like Japan's proposal, court rulings in the United States, or regulatory frameworks in Europe, the AI industry is moving toward a world where training data has a price.

The question is no longer whether AI companies will pay for creative works, but how much — and Japan may be the country that sets the benchmark. For the estimated 2 million professional creators in Japan alone, this legislation represents more than policy. It represents recognition that human creativity has value that technology must respect, not simply consume.