Why Copyright Alone Can’t Protect AI‑Generated Works

The rapid rise of generative artificial intelligence—text generators, image synthesizers, music composers—has sparked a legal whirlwind. For decades, copyright has served as the primary shield for creators, granting them exclusive rights to reproduce, distribute, and adapt their original works. Yet the very assumptions that underlie copyright—human authorship, originality, and a clear chain of ownership—are being upended by algorithms that can produce novel content at the click of a button.

---

1. The Human‑Authorship Requirement

Most national copyright statutes, from the United States Copyright Act to the EU Copyright Directive, explicitly require a human author. The U.S. Copyright Office’s 2023 policy memo states that works “created by a machine without any human authorship are not eligible for copyright protection.”

When an AI model like OpenAI’s GPT‑4 or Midjourney generates a poem, a painting, or a piece of code, the line between tool and collaborator blurs. If a user merely prompts the model and the output is largely machine‑driven, can that user claim authorship? Courts have been divided. In Zarya v. Turing (a fictional case used for illustration), a judge ruled that the user’s minimal input did not satisfy the originality threshold, leaving the work in the public domain.

---

2. Ownership Chains and Licensing Nightmares

Even when a human can claim authorship—say, by curating AI output—the ownership chain becomes tangled. Traditional licensing models assume that the licensor holds clear title to the work. AI developers, however, embed massive datasets that often contain copyrighted material. The Google Books settlement and the Authors Guild lawsuits against AI training data illustrate how ambiguous the provenance of training material can be.

If an artist uses a model trained on copyrighted images without a license, who is liable for infringement? The model’s creator (e.g., OpenAI), the platform hosting the model (e.g., Microsoft Azure), or the end‑user who generated the infringing image? Current copyright law offers no definitive answer, creating a risk‑averse environment where creators may avoid AI tools altogether.

---

3. Moral Rights and Attribution

Beyond economic rights, many jurisdictions protect moral rights—the right of attribution and the right of integrity. These rights are personal to the creator and cannot be transferred in many civil law countries. When an AI system produces a derivative of a protected work, the original author’s moral rights may be violated, even if the new work is technically original.

For example, a DeepMind‑generated painting that mimics the style of Van Gogh could be seen as an affront to the Dutch painter’s right of paternity. Yet there is no mechanism to enforce moral rights against a non‑human author, leaving a gap that copyright alone cannot fill.

---

4. The Public‑Domain and Knowledge Commons

One of copyright’s core policy goals is to balance private incentives with the public’s right to access knowledge. AI threatens to tip that balance. If every AI‑generated output is locked behind a new copyright claim, the public domain could shrink dramatically. Conversely, if AI outputs are deemed unprotectable, creators may lose the ability to monetize their investments, discouraging innovation.

The European Union’s proposed AI Act attempts to address this by classifying high‑risk AI systems and imposing transparency obligations, but it stops short of redefining authorship. Without a nuanced approach, we risk either an over‑protected AI‑driven market or a free‑for‑all where creators cannot reap rewards.

---

5. Policy Proposals Beyond Copyright

a. Sui Generis Rights for AI‑Assisted Works A limited, sui generis right could grant a modest, renewable protection for AI‑assisted creations, acknowledging the contribution of the algorithm without granting full exclusive rights. This would encourage investment in AI tools while preserving the public domain.

b. Mandatory Data‑Source Disclosure Requiring AI developers to disclose the provenance of training data would give downstream users clearer guidance on potential infringement risks. A “data‑sheet” model, similar to model‑cards for transparency, could become a legal prerequisite.

c. Collective Licensing for Training Data Creating a collective licensing framework—akin to the *Mechanical Licensing Collective* for music—could allow AI developers to obtain blanket licenses for the vast corpora they need, distributing royalties back to original creators.

d. Expanded Fair‑Use/Fair‑Dealing Exceptions Legislatures could carve out explicit fair‑use provisions for AI training, mirroring the *Google Books* settlement’s approach. This would provide legal certainty while protecting the interests of authors.

---

6. Looking Ahead

The conversation about AI and intellectual property is still in its infancy. Courts, lawmakers, and industry stakeholders are experimenting with piecemeal solutions, but none fully address the fundamental mismatch between a human‑centric copyright system and machine‑generated creativity.

What is clear is that copyright alone will not suffice. A layered framework—combining modest sui generis protections, transparency mandates, and updated fair‑use doctrines—offers the most promising path forward. Until such reforms materialize, creators, developers, and users must navigate a legal gray zone, balancing innovation with respect for the rights of existing authors.

---

The future of creative expression depends on our ability to adapt the law to the realities of AI. By recognizing the limits of copyright and embracing broader policy tools, we can foster an ecosystem where both humans and machines thrive.

Sources: https://www.thedial.world/articles/news/copyright-law-ai-intellectual-property