AI and Copyright: What Researchers Should Know

Researchers often ask: How does copyright law apply to the use of generative AI platforms?

The answer: Laws applying to generative AI and copyright are still in flux.

Nations across the globe are wrestling with how to regulate AI to balance the interests of copyright holders, who are concerned with protecting their moral and economic rights, and AI companies, for which profits can be maximized by using copyrighted works without consent or payment.

These questions are answered at the level of national law and policy. Each of the nearly 200 nations of the world sets its own copyright laws that apply within its jurisdiction. Numerous governmental consultations on AI have been conducted, resulting in reports, proposals, and policy documents, and in some cases, new laws. Eventually the world’s laws will address AI and copyright more clearly. In the meantime, lawsuits have been filed against AI companies to resolve disputes in the absence of legal clarity. As of June 22, 152 AI lawsuits had been filed globally, of which 118 were in the United States.

What This Means for Researchers

The more researchers understand the current landscape of copyright and generative AI, the better they will be able to navigate the terrain and avoid future challenges. Understanding the directions being taken to address AI nationally and internationally is critical.

Countries around the world are weighing various options for regulation of AI-generated content and training of large language models:

• Copyright exceptions for text and data mining (TDM) are often framed as “research” exceptions, but they are also intended to support the growth of AI. They are heavily promoted by the AI industry as a legal mechanism to avoid paying for the use of copyrighted works. Rightsholder groups oppose TDM exceptions, viewing them as a loophole that shuts them out of revenues. Some countries, including Japan, Singapore, and the United Kingdom, have TDM exceptions. TDM exceptions have been proposed in Hong Kong and Korea. They do not exist as such in U.S. copyright law.
   
• Compulsory licenses mandate use of content for AI training in return for small government-regulated payments to rightsholders. Rightsholder groups, as well as AI developers that are not interested in paying for training content, generally oppose compulsory licenses. There have been proposals for compulsory licensing in the European Union and in India, but they have faced strong opposition. The U.S. does not have a compulsory license for AI.
   
• Voluntary licenses benefit both the content and AI industries because they are based on agreement and mutual cooperation. Presently, the EU and the UK are proposing stronger licensing frameworks to restore the bargaining power of rightsholders and ensure viable protection. Licensing solutions have recently expanded, and AI licensing markets continue to grow.
   
• The “fair use” limitation is favored by U.S. technology giants claiming that the creation of large datasets to train AI models is “transformative” and qualifies as fair use. Like the TDM exception, “fair use” shuts rightsholders out of economic gains from AI. The “take first, claim fair use later” strategy also compromises rightsholders by forcing them into legal battles that many do not have the means to wage. On March 20, the U.S. government released ;A National Policy Framework for Artificial Intelligence expressing the administration’s belief that training of AI models using copyrighted works “does not violate copyright laws” and encouraging continued reliance on the judicial system to resolve fair use cases. This is an indication of the government’s policy preference for fair use as a legal strategy that supports the technology industry.
   

If you are using an AI platform in the U.S., you will be subject to U.S. law unless the end-user license states otherwise. Other aspects to consider are:

• Copyright protection for AI-generated works: In the U.S. it has long been held that copyrighted works must be created by a human being. In works consisting of both human-authored and AI-generated material, copyright may be claimed in eligible human-authored contributions. Whether a work created with the assistance of AI is copyrightable depends on the degree and nature of human involvement, and this is a case-by-case assessment. By contrast, a small number of countries provide protection to certain types of computer-generated works where there is no human author. They include China, the UK, Ireland, South Africa, and Ukraine.
   
• Copyright ownership in AI-generated works: In the U.S., to the extent that copyright may be claimed in eligible human-authored contributions in a work that consists of both human-authored and AI-generated material, copyright ownership in the human-authored contributions is held by the author, or in a work-made-for-hire scenario, by the employer or commissioning party, unless otherwise stipulated.
   
• Contracts or licenses applying to AI-generated works: It is important to be familiar with any contracts or licenses that apply. For example, some online licenses, or “terms of use” within AI platforms, assign “ownership” of AI outputs to users. One should keep in mind that depending on the degree of human authorship, there may be no copyright to be “owned.”
   
• Liability for creation and use of AI-generated works using AI platforms. AI-generated content is potentially subject to claims of copyright infringement if there is substantial similarity of the output with a copyrighted work and if it can be proven that the underlying work was actually copied. Additional risks may involve copyrighted fictional characters appearing in an output; digital images, videos, or audio recordings that realistically but falsely depict an individual (digital replicas); and AI outputs that simulate the artistic style of a human creator.
   
• State laws related to AI. In addition to federal copyright law, many states in the U.S. have adopted legislation on various aspects of law related to AI such as deepfakes, digital replicas, AI in political advertising, social media and online platforms, AI liability, and AI in education. New Jersey has adopted state laws on user-facing AI, automated decision-making, AI deepfakes, and AI in employment.
   

How Rutgers University Libraries Can Help

It is important that people using generative AI tools have basic knowledge of copyright and other legal issues surrounding their activity. Rutgers University Libraries have librarians with expertise in this area, and we can direct you to resources to help you acquire this knowledge. For questions about copyright, licensing, and intellectual property, send an email to ask_research@libraries.rutgers.edu.

This information does not constitute and should not be interpreted as legal advice, and it does not reflect any policy or practice of Rutgers University.