AI Content Licensing in 2026: Where the Major Cases Stand

The legal battles over AI training data have been running long enough that their current status is genuinely hard to track. Cases that launched with high-profile filings in 2023 and 2024 are now in discovery, in appeals, or in quiet settlement discussions that produce little public information. New filings have added to the docket. And a handful of licensing deals struck outside of court have begun to establish what voluntary arrangements between content owners and AI developers actually look like.

This is a field-wide status update, not a prediction. Courts are unpredictable. Settlement negotiations are private. But understanding where the major disputes stand is useful for anyone building on AI systems whose training data provenance is in question.

The Getty vs. Stability AI Cases

Getty Images filed against Stability AI in both the UK and the United States, arguing that Stability's use of Getty's photo library to train Stable Diffusion constituted copyright infringement at scale. The UK filing came first, in January 2023. The US filing followed in February of the same year.

The UK and US cases have proceeded on different timelines under different legal frameworks. In the UK, the case has moved through procedural stages under laws that treat training data use somewhat differently than US copyright doctrine. In the US, the case sits in the broader landscape of ongoing judicial interpretation of what fair use means when applied to AI training.

As of mid-2026, neither case has reached a final verdict. Discovery has been extensive. The evidence gathering around what data was used and how has produced public information about Stability's data practices that was not previously disclosed. Settlement discussions have been reported at various points but have not produced announced agreements. Both sides appear prepared to continue litigating.

The significance of the Getty cases extends beyond the immediate parties. A verdict that training on licensed or unlicensed copyrighted images constitutes infringement would affect the entire image generation industry. A verdict that finds fair use applies to training would provide a legal shield that AI developers have been seeking. Either outcome would reshape how AI companies approach data acquisition. The uncertainty itself has already had effects: many AI developers have taken more careful approaches to training data documentation and licensing than they might have otherwise.

The New York Times and the OpenAI Cases

The New York Times sued OpenAI and Microsoft in late 2023, alleging that training on Times content and the reproduction of that content in outputs crossed copyright lines in ways that fair use does not protect. The case was significant for its specificity. The Times' complaint included examples of GPT-4 reproducing Times articles at length, evidence of a kind that is more concrete than arguments about training data use in general.

The case has proceeded through its early stages with both sides filing extensive briefs. OpenAI and Microsoft's defenses have centered on the fair use argument and on challenges to the Times' characterization of how the models use training data. The Times has continued to build its factual record around reproduction cases.

Other publishers have filed separate but related complaints. The field of news publisher litigation against AI developers broadened significantly through 2024 and 2025. Some publishers struck licensing deals rather than litigating; others filed separately. The combination of litigation and negotiated licensing means that the ecosystem around news content and AI training is being reshaped on multiple tracks simultaneously.

What a resolution in the Times case would mean depends on which direction it goes. A ruling favorable to the Times would likely accelerate licensing negotiations across the news industry and could affect the economics of large language model training substantially. A ruling favorable to OpenAI would reduce the use that content owners currently have in licensing discussions, though it would not eliminate it entirely since reputational and relationship concerns also drive deals.

The case remains unresolved as of this writing, with significant procedural stages still to come.

The RIAA Cases Against Suno and Udio

The Recording Industry Association of America filed against AI music generation companies Suno and Udio in 2024, alleging that these companies trained their models on recordings in ways that infringed copyright. The music industry had been watching the image and text cases and moved more quickly to litigation than some observers expected.

The Suno and Udio cases raised questions that are somewhat different from the image and text cases. Music copyright involves both composition rights and recording rights, with separate holders and separate legal frameworks governing each. Whether training on recordings that contain both composition and performance elements creates infringement claims at the composition level, the recording level, or both is a question the courts have not clearly answered.

Both companies disputed the characterizations in the RIAA filings and defended their practices as falling within fair use. The cases have been in active litigation, with filings, responses, and procedural developments that are difficult to summarize briefly without misrepresenting the current posture.

The music industry's interest in controlling AI training data is partly about copyright protection and partly about the competitive threat that AI music generation poses to the revenue streams of labels and artists. This dual motivation, legal protection and market protection, shapes how the industry approaches settlements. A licensing arrangement that requires AI music companies to pay for training data access also creates a cost structure that raises barriers to entry and limits competition with the industry's own products. Whether the courts or legislative action eventually force the music industry to accept a fair use interpretation of training would significantly change the economics of AI music generation.

Licensing Deals Outside the Courts

Alongside the litigation, a parallel track of voluntary licensing has developed. Several AI companies have struck deals with content owners that provide cleared rights to training data in exchange for fees, revenue sharing, or other arrangements. These deals are rarely disclosed in detail, but their existence has become widely known.

The AP's arrangement with AI companies, deals between music publishers and AI developers, and licensing discussions between stock photo providers and image model developers have all been reported to varying degrees. The terms are generally confidential. The pattern is that content owners with sufficient use, either because of the quality of their content, their legal position, or their market influence, have been able to negotiate paid arrangements.

What these deals establish, over time, is a market expectation about what licensed training data costs. If a meaningful portion of AI training data is governed by licensing agreements with known terms, those terms begin to function as industry pricing signals. AI companies that trained before these norms emerged operate under different cost structures than those building new models today. This creates a form of historical advantage for early movers that has its own competitive implications.

The Broader Legal Uncertainty

The consistent thread across all of these cases is uncertainty. Courts have not yet produced definitive rulings on the key questions: whether training on copyrighted material constitutes infringement under current law, whether outputs that reproduce copyrighted content create separate claims from training-related claims, and how fair use applies to AI training at the scale and commercial purpose that characterizes current AI development.

This uncertainty is not neutral in its effects. It imposes legal risk on AI developers that favors well-capitalized companies that can sustain litigation exposure. It creates incentives to document data provenance more carefully. It accelerates the development of synthetic training data alternatives that sidestep copyright questions. And it keeps content owners at the negotiating table when they might otherwise have less use.

Legislative activity in multiple jurisdictions has added another layer of complexity. Proposed rules and regulations around AI training data transparency, opt-out mechanisms for copyright holders, and minimum licensing requirements have moved through various stages in different countries. None of these have produced final law that substantially resolves the underlying questions, but the possibility of legislative intervention affects the incentives of all parties in ongoing litigation.

The resolution of these cases, whenever it comes, will determine the legal architecture that governs AI training data for the foreseeable future. That architecture will shape what can be built, at what cost, and by whom.