Intellectual Property in the Age of Generative AI

The legal frameworks that govern intellectual property were built on a foundational assumption: that creation requires a human creator. Copyright, in particular, rests on the romantic premise that authorship reflects individual expression — a spark of human originality that the law then protects as property. Generative AI does not fit this premise. It does not fit because it was never anticipated.

The result is a period of extraordinary legal uncertainty that creates both risk and opportunity for sophisticated enterprises.

The Training Data Question

Every large language model, image generator, and code synthesis system is trained on data. That data — text, images, code, audio — was created by humans, frequently under copyright. The fundamental question that courts across the United States, United Kingdom, and European Union are now being asked to answer is whether training a model on copyrighted data constitutes infringement.

The argument for infringement: training requires the creation of a compressed, derivative representation of the original work. That representation is retained in the model’s weights and influences its outputs. The model has, in a technical sense, learned from the work without license.

The argument against: the model does not reproduce the original work. It produces statistically likely continuations of learned patterns. The analogy to human authors — who also learn from existing works — is imperfect but not irrelevant.

Current U.S. jurisprudence has not resolved this question. The Andersen v. Stability AI litigation, Getty Images v. Stability AI, and a cluster of cases in the Northern District of California are working through these issues, but final resolution at the appellate level remains years away.

What this means for companies deploying generative AI: the training data provenance of every AI system they rely upon is a legal liability of indeterminate magnitude. Counsel who is not conducting training data audits on AI vendor agreements is leaving their clients exposed to a claim that may be worth nine figures.

Output Ownership: Who Holds the Copyright?

If a human author uses generative AI as a tool — much as a photographer uses a camera — the human creative contribution may be sufficient to ground copyright ownership. If, however, the AI system generates output with minimal human creative direction, the U.S. Copyright Office’s emerging position is that no copyright subsists in the output.

This creates a paradox for companies building AI-generated content at scale. The content they are generating — product descriptions, marketing copy, design elements, code — may be unprotectable as intellectual property. It can be copied by competitors without license. It can be reproduced without attribution or compensation.

The strategic response is to re-engineer the creative workflow to maximize documented human contribution. Not as a fiction — but as a genuine integration of human creative judgment into the generative process. Prompt engineering is not a commodity skill; it is, increasingly, an IP strategy.

Patent Protection for AI-Generated Inventions

The patent landscape for AI-generated inventions is similarly unsettled. The Federal Circuit’s decision in Thaler v. Vidal confirmed that only human beings can be listed as inventors on U.S. patent applications. AI systems cannot be inventors.

But the question of whether AI-assisted invention — where the AI substantially contributes to the inventive concept while a human directs the inquiry — satisfies inventorship requirements is a live issue. The answer depends on the nature of the human contribution, the documentation of the inventive process, and the sophistication with which the application is prosecuted.

For companies in technology, pharmaceutical, and materials science sectors, the inventorship question is not theoretical. Laboratory AI systems are identifying novel molecular configurations, code synthesis systems are producing patentable algorithms, and materials discovery platforms are generating compositions that no human chemist independently conceived. The human scientists who directed those systems, reviewed their outputs, and reduced them to practice are inventors. But proving that requires contemporaneous documentation that most companies are not generating.

Trade Secrets as the Practical IP Regime for AI

Where copyright is uncertain and patent requires disclosure, trade secret protection has emerged as the most practically valuable IP regime for AI-era innovation. Trade secret protection is immediate upon creation, requires no registration, extends indefinitely as long as secrecy is maintained, and covers exactly the information that makes AI systems valuable: training data, model architectures, fine-tuning techniques, and prompt libraries.

Effective trade secret protection requires reasonable measures to maintain secrecy. For AI systems, this means access controls on model weights, confidentiality obligations for employees with system access, vendor agreements with strict use limitation provisions, and comprehensive security protocols that can be documented and produced in litigation.

The Defend Trade Secrets Act provides federal jurisdiction for trade secret claims, making national enforcement practicable. We have secured injunctions and substantial damages awards under the DTSA for clients whose AI model weights were misappropriated by departing employees and by vendor partners who exceeded their licensed access.

Conclusion: Establish Ownership Before Ownership Matters

The window in which AI-era IP frameworks are being established is closing. Courts are developing doctrine. Legislatures are drafting statutes. Regulatory bodies are asserting jurisdiction. Companies that are building comprehensive IP strategies now — that understand what they own, what they have licensed, what they have created, and what protections apply — will be positioned to enforce those rights when the frameworks stabilize.

Companies that wait will find the foundational questions already answered — by someone else’s litigation, in someone else’s favor.