US Pushes G20 to Back Fair Use for AI Training on Copyrighted Work

The United States has asked G20 countries to develop rules that permit artificial-intelligence companies to train models on copyrighted books, music, art and other creative work while preserving protections for creators. The push places a contested interpretation of copyright—fair use—at the center of a broader call for lighter-touch AI governance. 1

What the US asked G20 countries to do

At a technology-focused G20 meeting in North Carolina, Commerce Secretary Howard Lutnick urged governments to embrace frameworks that allow AI training on creators’ work. His argument was that countries should protect artists and inventors without erecting barriers that prevent innovation. 1

Lutnick specifically invoked fair use, the US copyright doctrine that can allow certain unauthorized uses of protected work. But the proposal did not spell out how countries should deliver both sides of that balance. It left unanswered practical questions such as when permission would be required, whether creators would be paid, and how protections would be enforced. 1

That omission matters because copyright rules vary by jurisdiction, and the legality of using copyrighted material for model training remains actively disputed.

Why AI training data is in court

Anthropic, OpenAI, Google and Meta have faced lawsuits from authors, publishers, media organizations and other creators over the alleged use of copyrighted material in AI development. The claimants contend that their work was used without authorization; the companies have argued that the training process is lawful. 1

The US Justice Department took a notable position in OpenAI’s dispute with The New York Times and other newspapers, telling a court that AI training generally constitutes fair use of copyrighted material. 1

That is a legal position, not a final resolution of every case. Whether fair use applies is assessed through litigation and can depend on the facts of a particular use. The mounting cases therefore remain consequential for the AI industry and for creators seeking control over how their work is used.

A broader case for regulating real-world harms

The copyright discussion came alongside a wider US argument against sweeping new AI rules. Nvidia CEO Jensen Huang told G20 officials that governments should avoid regulations aimed at “theoretical harms” and instead address concrete, real-world problems associated with the technology. 1

The distinction reflects a policy preference for applying rules to demonstrated risks rather than creating broad, technology-specific restrictions in advance. Supporters say that approach can preserve innovation; critics of a hands-off model may argue that waiting for harm can leave gaps in protection. The G20 discussions did not settle that underlying policy disagreement.

What the G20 statement said about future AI rules

The G20 joint statement said future AI-specific regulation should be limited to “novel considerations” that existing laws do not already address. China signed the statement despite already having broad AI regulations. 1

The policy direction was tied to concerns that extensive government requirements could affect the industry’s economics and product roadmaps—for example, if they delay new model releases or require companies to change how products operate in response to security demands. 1

The unresolved trade-off

The G20 discussion points toward a policy compromise: preserve room for AI developers to use training data under fair-use principles, but protect the creators whose work helps make those systems possible. The hard part is still unresolved. Without clear rules for licensing, compensation, consent or enforcement, the balance Lutnick called for will continue to be shaped by national law and ongoing copyright litigation. 1