AI Copyright Law in 2026: What Courts and Lawmakers Decided

AI copyright law in 2026 has developed more quickly than most predicted, driven by litigation from artists, authors, publishers, and record labels that forced courts to make decisions before legislators caught up. The legal landscape is still evolving, but several significant rulings and legislative developments have clarified some of the most contested questions — and created new ones.

This overview focuses on where the law currently stands, what it means for AI developers, and what remains unresolved.

The Core Question: Does Training AI on Copyrighted Work Require Permission?

The central dispute in AI copyright law has been whether using copyrighted text, images, music, and code to train AI models constitutes copyright infringement.

The leading US theory in favor of AI developers is fair use. The argument: training is transformative (the model doesn't reproduce the input data, it learns statistical patterns), using works for research and development purposes is consistent with fair use policy, and the market impact is distinct from the original works.

The argument against: copyright holders never licensed their work for this purpose, AI outputs can substitute for licensed content in ways that reduce demand for originals, and the scale of copying involved is unprecedented.

Courts have not issued a single definitive ruling. Multiple cases have settled or produced narrow rulings, and circuit splits are emerging. The Supreme Court is widely expected to eventually take a case that could establish clearer doctrine, but that has not happened as of September 2026.

Significant Court Rulings in 2025-2026

Several cases have produced meaningful precedents and settlements:

The visual art class actions. Multiple class action lawsuits by visual artists against image generation companies have produced settlements that, while not admitting liability, have resulted in large payouts and opt-out mechanisms for artists. Importantly, none of these cases produced a published opinion establishing infringement — they settled before the key questions were judicially resolved.

The news publisher licensing agreements. Following the lead of the AP's licensing agreement with OpenAI in 2023, most major US news publishers have now signed licensing agreements with AI companies for training data use. Whether these agreements reflect a legal requirement or a business negotiation strategy is debated — but the practical result is that AI companies are now paying for text data from major publishers.

Copyright in AI outputs. US courts have consistently held that purely AI-generated content cannot receive copyright protection because copyright requires human authorship. The Copyright Office's guidance through 2025 and 2026 has reinforced this position while acknowledging that AI-assisted human creative work can be protected when the human contribution is meaningful and identifiable.

Code copyright disputes. GitHub Copilot litigation over whether AI-generated code that closely resembles training data infringes on the original code's copyright has produced mixed lower court rulings. The cases have highlighted the difficulty of applying traditional copyright concepts to statistical patterns in code generation.

Legislative Developments in 2026

Several countries have enacted or proposed AI-specific copyright legislation:

EU approaches. The EU AI Act includes transparency requirements for GPAI models about training data provenance. Separately, the Digital Services Act framework has been interpreted by some member states to impose additional obligations around AI training data. The EU has not yet passed specific legislation creating a training data copyright exception or establishing a remuneration right.

UK consultation. The UK government launched a consultation on AI and copyright in 2024-2025, proposing a text and data mining exception that would allow AI training on legally accessed content. As of September 2026, final legislation has not been enacted, and the proposal remains contentious with the creative industries.

US federal stalemate. Congress has held numerous hearings on AI and copyright but has not passed federal legislation. Bills have been introduced that would require disclosure of training data, establish licensing requirements, and create creator compensation funds, but none have advanced. The lack of federal legislation has increased the importance of judicial decisions and state-level activity.

What This Means for AI Developers in 2026

The practical AI copyright law situation for developers in 2026 is characterized by uncertainty with some reliable rules:

AI output is not copyrightable without human authorship. Content produced purely by AI cannot be copyrighted by the AI developer or the user. Users who significantly direct, edit, or otherwise make creative choices in AI-assisted work may retain copyright in those human contributions.

Training data provenance matters. For enterprise and commercial deployments, demonstrating that training data was acquired through legitimate licenses or legally permitted means has become important for risk management even where the law remains unsettled. Major AI companies are building training data provenance documentation as a standard practice.

Opt-out mechanisms and licensing are becoming standard. The market is moving toward AI companies respecting creator opt-out signals and offering licensing deals for large-scale training data use. Companies that ignore creator opt-out signals face both legal and reputational risk.

Watermarking and disclosure norms are developing. While no federal disclosure requirement exists in the US, norms and some state-level requirements are developing around disclosing AI-generated content in specific contexts (journalism, legal filings, academic work). Keeping current with the disclosure requirements in specific contexts is necessary for professional use.

For the related legal risk picture, the AI legal liability in 2026 overview covers how liability frameworks are developing alongside copyright doctrine.

The Creative Industries' Response

The creative industries have responded to AI copyright uncertainty through several channels:

Collective licensing negotiations. Artists, authors, and music rights holders have increasingly organized to negotiate licensing terms with AI companies, using their leverage as data providers. Some licensing pools have been established specifically for AI training.

Technical tools for creator protection. Services that allow creators to add signals to their digital content that AI training systems can detect and respect have attracted significant investment. The practical effectiveness of these tools depends on AI developers actually honoring the signals, which remains voluntary absent legislation.

Portfolio repositioning. Some creators and companies are repositioning their IP strategy around the AI era — focusing on what AI can't easily produce (physical craft, live performance, relationship-based services), pursuing licensing income from AI companies, or building AI tools themselves.

The International Dimension

AI copyright law in 2026 is highly fragmented internationally. Japan has taken a permissive approach, explicitly permitting AI training on copyrighted works. The EU is taking a more restrictive path. The US remains in a judicial and legislative gray zone.

This fragmentation creates practical challenges for globally deployed AI systems. A model trained in one jurisdiction may face different legal treatment when the outputs are used in another. The trend toward AI companies maintaining provenance records about training data is partly a response to this jurisdictional complexity.

What's Still Unresolved

Several major questions in AI copyright law remain open in September 2026:

• Whether large-scale AI training constitutes fair use under US law (no appellate court has squarely addressed this)
• What level of similarity between AI output and training data constitutes infringement
• How licensing requirements would work in practice if mandated by legislation
• Whether creators have a right to compensation when their work is used in training, even if the training is ultimately found to be legal
• How the AI Act's transparency requirements will be enforced in practice regarding training data disclosure

The next 12-18 months are likely to produce more clarity on at least some of these questions, either through court rulings or legislative action in the EU or UK. For AI developers and businesses relying on AI, monitoring the legal landscape and documenting training data provenance are the most immediately practical risk management strategies available.