Copyright Protection for AI Outputs: The Human Authorship Requirement and Emerging Case Law
The surge of generative AI has pushed copyright law into unfamiliar terrain. A central question is whether a work produced by an AI can receive copyright protection. In the United States, copyright attaches to original works of authorship fixed in a tangible medium. The pivotal criterion is human authorship—a condition that courts and the U.S. Copyright Office have consistently enforced.
This human authorship threshold requires that a work stem from a human mind and exhibit at least a minimal degree of creativity. The principle was notably examined in Naruto v. Slater (the "monkey selfie" case), where the Ninth Circuit ruled that a macaque could not bring a copyright infringement suit because animals do not have statutory standing under the Copyright Act. In a parallel vein, Thaler v. Hirshfeld saw the Federal Circuit confirm that an AI system cannot be named as an inventor on a patent, underscoring the human-centric orientation of intellectual property law.
More recent decisions shed further light on the boundaries. In Andersen v. Stability AI, artists filed suit against Stability AI, Midjourney, and DeviantArt for copyright infringement, claiming that the AI models were trained on their works without authorization. The case is still pending, but it brings the friction between AI training and copyright into sharp relief. The U.S. Copyright Office has also refused registration for works generated solely by AI, stressing that human creativity is essential.
Training Data and Fair Use: Litigation and Licensing Approaches for Creative Works
Using copyrighted works as training data for AI models has ignited fierce legal disputes. The fair use doctrine, set out in Section 107 of the Copyright Act, permits limited use of copyrighted material without permission for purposes such as criticism, comment, news reporting, teaching, scholarship, or research. Courts weigh four factors: (1) the purpose and character of the use, (2) the nature of the copyrighted work, (3) the amount used, and (4) the effect on the potential market.
In Authors Guild v. Google (Google Books), the Second Circuit concluded that Google's digitization of books for its search engine was transformative and qualified as fair use. Yet the Supreme Court's ruling in Andy Warhol Foundation v. Goldsmith narrowed the transformative use doctrine, stressing that the commercial character and the specific use are important. These precedents are now being applied to AI training. Recent lawsuits, including those brought by Getty Images against Stability AI, allege that AI companies reproduced copyrighted images without authorization. The outcome will depend on whether training is considered transformative and whether it harms the market for original works.
To address litigation risks, licensing models are taking shape. Collective licensing initiatives, such as those by the Copyright Clearance Center, seek to offer a streamlined way for AI developers to license content. Adobe's Firefly model, trained on licensed and public domain content, provides an example of compensating creators. The EU AI Act also imposes transparency and copyright-related obligations on general-purpose AI systems, indicating a regulatory trend toward balancing innovation and creators' rights.
Ownership and Contractual Allocation Among Users, Platforms, and Model Providers
In addition to copyrightability and fair use, contracts often govern the ownership of AI-generated works. Platform terms of service and licensing agreements allocate rights among users, platforms, and model providers. For example, OpenAI's terms of use state that users own the output generated by its models, but this ownership is conditioned on compliance with the terms. Midjourney's terms grant users a license to use the images they create, while the platform retains certain rights. Adobe Firefly, by contrast, offers commercial licensing for its outputs, giving business users clarity.
These contractual arrangements carry legal uncertainties. The enforceability of ownership clauses may be challenged, particularly if the AI output incorporates third-party copyrighted material. Best practices for allocation include clear and conspicuous terms, warranties regarding the training data, and indemnification clauses. Comparative approaches in the EU and US diverge: the EU's AI Act emphasizes transparency and risk management, whereas the US relies more on existing copyright law and contract principles. Industry recommendations suggest that stakeholders should engage in proactive licensing and contractual clarity to mitigate risks.
Conclusion: Charting the Future of AI and Copyright
The intersection of AI and copyright forms a complex legal landscape. The human authorship requirement remains a cornerstone, but its application to AI-generated works is still developing. Fair use in training data is hotly contested, with litigation and licensing models shaping the future. Contractual allocation of ownership offers some certainty, yet legal uncertainties persist. As generative AI continues to advance, legislative reforms and international harmonization will be essential to foster innovation while protecting creators' rights. Stakeholders must stay informed and adapt to this rapidly changing environment.
