Six pillars in a new Group of 20 innovation statement place intellectual property alongside artificial intelligence policy, turning a technical debate over model training into a direct economic question for studios, record labels, publishers and working creators. The consensus text released Wednesday recognizes copyright's role in sustaining human creative work, but leaves each country to decide when consent, exceptions and compensation apply.

That carefully balanced language followed a sharper American message. Commerce Secretary Howard Lutnick urged governments to let AI companies train on creative works under fair-use principles while also protecting artists, according to Reuters. He offered no mechanism for deciding which uses would be fair or how creators would be paid. For entertainment businesses, that omission matters more than the diplomatic agreement: the value of a film library, music catalog or script archive increasingly depends on who may copy it into a training system and on what terms.

The meeting therefore produced a meaningful policy signal, not a global license. It aligned major economies around adaptable rules and existing legal institutions while preserving national sovereignty. It did not settle whether unlicensed ingestion is lawful, create a payment system or bind courts considering pending cases.

What the G20 Actually Agreed

The White House said ministers meeting in Chapel Hill, North Carolina, reached consensus on the Carolina Principles and a broader innovation statement. Participants included the European Union, China, Britain, Japan, India, Brazil and other major economies. The principles encourage investment in foundational research, faster commercialization and reliance on existing sector-specific regulation unless an emerging technology creates a genuinely new problem.

The copyright section is more qualified than the administration's public advocacy. The ministerial text says intellectual-property frameworks can support AI developers and creators, recognizes that copyright safeguards human work, and notes that countries use different doctrines, including express consent and copyright limitations or exceptions. It also supports fair remuneration and remedies against misappropriation under applicable law.

That combination gives every side language it can cite. AI developers can point to flexible policy and established exceptions. Rightsholders can point to consent, remuneration and enforcement. Because the statement directs disputes back to domestic legal processes, its near-term effect will be persuasive rather than compulsory. The real rules will still be written through lawsuits, licenses, labor contracts and national legislation.

The administration reinforced its international position with a domestic court filing. The Justice Department submitted a brief supporting OpenAI in the New York Times' copyright case, arguing that training large language models is generally transformative and can qualify as fair use. As legal reporting noted, the brief is advisory and does not decide the dispute, but it appears to be the federal government's first intervention in the current wave of AI-training cases.

Fair use is not a blanket exemption. Courts weigh the purpose and character of a use, the nature of the protected work, how much was taken and the effect on actual or potential markets. A model used for research may present a different case from one designed to generate commercial songs, scripts or images that substitute for licensed work. The source of the training copy can also matter, particularly when plaintiffs allege that material came from pirate repositories.

That nuance is visible in the federal government's own expert record. A 2025 Copyright Office report concluded that some training uses are likely transformative, but said commercial use of vast troves of protected works to produce competing expressive content can exceed established fair-use boundaries, especially when the works were illegally accessed. It recommended allowing voluntary licensing markets to develop rather than imposing a compulsory system immediately.

Music and Film Face Different Exposure

Entertainment is not one rights market. Film and television productions divide control among studios, writers, performers, directors, composers and guild agreements. Recorded music separates sound-recording rights from composition rights. Books, journalism, photographs and visual art have different ownership patterns again. A global principle that simply invokes fair use cannot resolve those chains of title or tell a model developer whom to pay.

The conflict is already concrete in music. Sony and Warner publishing units recently accused Anthropic of copying hundreds of song lyrics and scores, seeking statutory damages and an injunction. Anthropic called the allegations recycled and said training is fair use. The music lawsuit raises two distinct questions: whether obtaining training material was lawful and whether a system's outputs can compete with the originals. A court could answer those questions differently even within the same case.

Hollywood's labor rules address another layer. The writers' contract says AI cannot write or rewrite literary material for purposes of writer credit, studios cannot require writers to use AI, and companies must disclose when assigned material contains AI-generated content. Those protections govern employment relationships; they do not determine whether older scripts can be used for model training. Copyright litigation and collective bargaining therefore operate in parallel, protecting different interests.

The International Split Remains

The G20 consensus does not erase national differences. The United States relies heavily on open-ended fair use and case-specific judicial decisions. The European Union's AI framework requires providers of general-purpose models to maintain a copyright-compliance policy and publish a sufficiently detailed summary of training content. The European Commission's AI overview describes tools for disclosing major datasets and source domains so rightsholders can exercise their rights.

Other jurisdictions use text-and-data-mining exceptions, consent rules or opt-out systems. Those approaches affect whether a film producer or music publisher must negotiate before training begins, may reserve rights after publication, or must challenge a use after it occurs. The G20 statement acknowledges this diversity rather than selecting a single legal test. That is diplomatically practical, but it leaves multinational entertainment companies managing several compliance regimes at once.

The divergence also shapes competition. Broad exceptions may lower entry costs for AI developers because they need fewer licenses, but they can weaken bargaining leverage for creators whose work supplies the training material. Strict consent rules can improve control and compensation, yet favor large rightsholders able to aggregate catalogs and strike deals. A legal analysis describes this as a double bind: creators may benefit from AI tools while fearing replacement and may not share equally in the value of corporate-owned rights.

The Next Decisions Are Commercial

For studios, labels and publishers, the immediate question is not whether the G20 authorized training; it did not. The practical work is catalog governance: identifying what a company owns, what it merely licenses, which contracts restrict reuse, and whether machine-readable reservations of rights will be honored. Provenance records become valuable because the legal analysis can turn on the source of a copy as much as the model's final output.

AI companies, meanwhile, have incentives to separate low-risk analytical uses from products that generate expressive substitutes, improve filters against verbatim reproduction and document how datasets were acquired. Voluntary deals can reduce uncertainty, but their economics will vary by sector. A major studio can negotiate a portfolio license; an independent illustrator or songwriter may need collective representation or standardized marketplaces to obtain similar leverage.

The Chapel Hill agreement moves the debate from whether copyright belongs in AI policy to how each country will apply it. Washington has now favored a broad innovation-oriented reading of fair use, while its own Copyright Office and creative industries emphasize source legality, market substitution and compensation. Until courts or lawmakers draw firmer lines, the entertainment economy will be governed by that tension—and by contracts written in its shadow.