Navigating the Copyright and AI Battleground

The landscape of intellectual property has been fundamentally disrupted. Generative artificial intelligence (AI) tools have altered the creativity process, enabling anyone with an internet connection to generate vivid digital artwork, write code, or draft essays in a matter of seconds.

Yet, this revolutionary technology collides head-on with 200 years of copyright jurisprudence. As technology outpaces the law, the creative and legal worlds find themselves in a heated battle. The conflict spans three critical areas: authorship, infringement, and enforcement.

The Non-Human Creator: Who Owns AI Outputs?

There are constitutional underpinnings to the human authorship requirement. The concept that creative agency belongs exclusively to human beings is a cornerstone of global intellectual property regimes. Generative AI has completely destabilized this architecture.

“[The Congress shall have Power … ] To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries.”

U.S. Constitution, Art. I, Sec. 8, Cl. 8

Legal precedent is unequivocal in that machines cannot create creative works nor can they own copyrights. In the United States, courts have repeatedly affirmed that copyright requires a human author. This principle was underlined by Naruto v. Slater,¹ the famous “monkey selfie” case (), where a macaque monkey was denied statutory standing as an author. Additionally, in Thaler v. Perlmutter,² the court of appeals affirmed the U.S. Copyright Office’s denial of registration for a creative work generated entirely autonomously by an AI system. The court made clear that a work lacking human involvement cannot be eligible for copyright registration.

The Gray Area of Human-Assisted Works

While purely autonomous AI outputs enter the public domain, the real world is rarely black and white. The U.S. Copyright Office issued updated guidance confirming that it will not refuse registration simply because a creator used AI as a tool. If a human exercises significant creative control—such as heavily editing the output, arranging elements into a unique compilation, or digitally manipulating the machine’s work—the human-authored components can be registered.

However, standard text prompts alone will rarely satisfy the legal requirement for human authorship. It leaves creators in a delicate position: use AI to enhance your work, but rely too heavily on the machine, and you risk losing all authorship, and along with it, the ability to register and own copyrights.

The Infringement Dilemma: Inputs vs. Outputs

The issue of copyright infringement regarding generative AI operates on two distinct fronts: upstream training (inputs) and downstream generation (outputs).

The Upstream Battle: Training Data and Fair Use

AI models do not learn in a vacuum; they require massive datasets consisting of billions of images, text strings, and musical tracks, often scraped from the open web without explicit consent.

Model developers argue that this process constitutes fair use under U.S. law, or falls under text and data mining exceptions globally. They argue that training is transformative use because the model learns underlying mathematical patterns to create entirely new works rather than regurgitating the originals.

Conversely, content creators, authors, and major publishers argue that this is unauthorized mass secondary infringement on an unprecedented scale.

While U.S. courts are still working through major fair use lawsuits, international courts are shifting toward stricter rules. For instance, in Europe, the GEMA v. OpenAI[3] ruling in Germany held that AI training created copyright-relevant reproductions that were not sheltered by standard data-mining exceptions, signaling a tougher regulatory approach abroad.

The Downstream Battle: Substantial Similarity

Infringement can also occur when an AI model outputs material that is substantially similar to a copyrighted work. If a user prompts a model to generate a movie poster featuring a copyrighted character or write a story exactly in the style of a living author using their specific characters, the output can directly violate the copyright holder’s exclusive rights.

Enforcement: How Do We Police the Infinite Machine?

Even as legal frameworks slowly solidify, enforcement remains an operational nightmare. AI models can generate millions of pieces of content per hour, creating a volume of potential infringement that completely overwhelms traditional legal remedies.

The Requirement of Registration

In the U.S., a copyright holder must officially register their work with the Copyright Office before they can file suit in federal court for statutory damages or attorney’s fees. However, for digital publishers and journalists who publish hundreds of articles daily, keeping up with standard registration protocols to combat AI, while statutorily necessary, can be burdensome, if not entirely impossible. (Outside copyright counsel is useful here.)

Technical and Legal Solutions Emerging

To curb unauthorized exploitation and enforce rights, the creative and legal industries are exploring new defense mechanisms, including utilizing specialized web protocols to explicitly block AI crawlers from scraping their sites, embedding digital watermarks and cryptographic signatures are being embedded into human-made creations to verify their origin and track unauthorized training usage, and negotiating licensing deals with tech companies, granting legal access to high-quality training data.

The Path Forward

The convergence of AI and copyright law represents one of the greatest intellectual property challenges of our generation. Human authorship will not change outside of a constitutional amendment. Working within this framework will advances in technology be able to co-exist with IP.

This blog posting is for informational purposes only. If you have a specific issue or question related to copyrights or AI, please contact Yonaxis I.P. Law Group.

---

Footnotes

1. 888 F.3d 418 (9th Cir. 2018). ↩

2. 130 F.4th 1039 (D.D.C. Cir. 2025). ↩