Adapting to AI: Copyright Law Will Catch Up, as it Always Has

“As AI technology improves and more products enter the market, Americans will continue to…enforce their rights in court. The results of those cases, along with decisions in currently pending matters, will inform the market as to the licenses it needs.”

Public discourse over the last few months illustrates the important role IP protections play in our society. Much of this has centered on the impact of new technologies, especially those powered by AI.

As new technologies enter the market, the government reacts. Before shaping reactionary IP regulation and legislation, however, lawmakers must understand both the technology and the rights it implicates. Then they must grapple with the great dilemma of IP law: how to promote innovation while still protecting earlier works. Fortunately, this process isn’t unique to AI. American IP law has adapted to numerous innovations over the years and can do so again now.

Learning from the Past: Copyrights on Computer Programs

Copyright law protects expressions of ideas from unauthorized copying. This made the application of copyright law to the first software programs unclear, since software serves as a functional expression of underlying ideas and often requires copies to function. In 1974, Congress responded to this lack of clarity by establishing the Commission on New Technological Uses of Copyrighted Works (CONTU). CONTU ultimately decided that computer programs were protectible forms of expression that users could permissibly copy in certain circumstances. Congress officially extended copyright protection to computer programs with the Computer Software Copyright Act of 1980. 17 U.S.C. § 101.

Learning from the Past: Web Host Liability

Once the internet became publicly available in the 1990s, people began placing their thoughts, ideas, and creative works on the web for others to see… and copy. Suddenly, copyrights became much easier to infringe. This became a major problem for anyone hosting a website—if they allowed people to post unmoderated content on their site and any of it infringed a copyright, the host would be liable. This created a no-win scenario for many web hosts: verifying the originality of each post would be impossible, but a single infringement suit could put them out of business.

Congress couldn’t react by removing all web-host liability though; if they did, anonymous users could infringe with impunity. So, they found a middle ground. To protect copyright holders while promoting innovation (i.e., websites with user-generated content), Congress passed the Digital Millenium Copyright Act (DMCA) in 1998. The DMCA provides a safe harbor for service providers hosting user-generated content subject to its “notice-and-takedown” requirements. 17 U.S.C. § 1201. Essentially, this meant that websites would only be liable for hosting infringing content they knew about.

Where We Are Now: Artificial Intelligence

Philosophically, the current debate over AI comes down to one fundamental question, as posed by Shira Perlmutter, former Register of Copyrights: “How do we respect and reward human creators without impeding technological progress?” (Report on Copyright and Artificial Intelligence, Part 1, Foreword). Practically, that breaks down into two novel issues of copyrightability: whether computer-generated content is copyrightable and whether using copyrighted material to train AI models constitutes infringement. The Copyright Office analyzed these issues in its recent reports of Copyright and Artificial Intelligence, Part 2: Copyrightability (January 2025) and Part 3: Generative AI Training (May 2025, Pre-Publication Version).

Copyrightability: Under our current copyright system, copyright ownership “vests initially in the author or authors of the work,” 17 U.S.C. § 201(a), and that author must be a “person.” Cmty. for Creative Non-Violence v. Reid, 490 U.S. 730, 737 (1989). The Copyright Office reiterated this position in Part 2 of its recent report, explaining that “[c]opyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.” But “where AI is used as a tool, and where a human has been able to determine the expressive elements they contain,” AI output can still be copyrightable.

In its report, the Copyright Office repeatedly emphasized what a fact-specific determination this is and, as such, provided multiple examples, including the AI-assisted work by artist Kris Kashtanova shown below. The report explained that Kashtanova received a limited copyright in their work, “with a scope analogous to that in a derivative work.” Whereas derivative work protection is limited to the material added by a later author, protection here was limited to the “perceptible human expression” in the final work, including the outline of the mask and the arrangement of facial features and foliage. But “any non-human expression,” including the shadows and the 3D flowers, nose, and lips, did not receive a copyright.

Click here to read the full article.