Intellectual Property (IP)

OVERVIEW

The collision between generative AI and established Intellectual Property frameworks, specifically Copyright, Patent, and Trade Secrets, represents the most active front in AI-related litigation. These frameworks were designed almost exclusively around human authorship and inventorship, creating profound legal friction.

Training Data Liability and Copyright

Large language and generative models (LLMs/LGMs) are trained on massive datasets scraped from the public internet, which often includes billions of copyrighted works. The central legal dispute, exemplified by actions against Stability AI, Midjourney, and OpenAI, hinges on the doctrine of Fair Use1 in U.S. Copyright law. 

AI developers argue that the act of copying for training is transformative, creating a new technical tool rather than a substitute for the original work, and that the resulting model weights do not reproduce the original expressive work. Conversely, content creators assert that unauthorized mass reproduction of their work, even for intermediate purposes, constitutes infringement, particularly because the AI output often competes directly with their original material. 

Litigants are seeking unprecedented damages and, critically, injunctive relief that could halt the distribution or require the costly destruction of models trained on infringing data. The outcome will fundamentally redefine the scope of copyright in the digital age.

Authorship and Ownership of AI Output.

Current U.S. and international IP law explicitly requires a human author for copyright registration. The U.S. Copyright Office has affirmed that it will register works only when a human has "creatively selected or arranged" the AI-generated elements, effectively ruling out AI as the sole legal author. This creates a significant ownership vacuum for purely autonomous AI creations, complicating commercialization and enforcement. 

A similar challenge exists in Patent Law, where the DABUS2 cases globally have affirmed the requirement for a human inventor, blocking AI systems from being formally recognized as such. The litigation challenge here is defining the necessary quantum of human input3 required to successfully claim ownership, a threshold that remains poorly defined for prompt engineering, finetuning, and curation.

Trade Secrets

Trade Secrets are the default protection mechanism for the AI model itself. Companies guard the model architecture, training data composition, and unique hyper-parameters as critical proprietary assets.

Litigation often arises from employee poaching, where departing employees attempt to leverage knowledge of these secret ingredients for competitors, or through reverse engineering attempts. This choice of Trade Secret protection is often a strategic hedge against the potential invalidation risks and disclosure requirements of patenting, but it places a heavy reliance on contractual protections (NDAs, non-competes) and strong physical/digital security, all of which must be aggressively defended in court to maintain their legal standing. 

The current litigation landscape is rapidly shaping commercial norms, necessitating a shift from reactive defense to proactive data governance and licensing strategies.

DISCUSSION AND CASE LAW

Training Data Liability and Copyright

The central disputes in this area involve allegations that AI companies infringed copyrights by using protected works to train generative models without authorization, with defendants often invoking the fair use doctrine. Key cases include:

• The New York Times Company v. Microsoft Corp. and      OpenAI (S.D.N.Y. 2023, ongoing in 2025 with partial fair use rulings      favoring AI training as transformative in related decisions)
• Getty Images (US), Inc. v. Stability AI, Inc. (D.      Del. 2023, focusing on unauthorized scraping of images for Stable      Diffusion training, with 2025 developments on injunctive relief)
• Andersen v. Stability AI Ltd. (N.D. Cal. 2023, class      action by artists claiming output competes with originals, leading to      mid-2025 fair use partial victories for defendants)
• Authors Guild v. OpenAI (S.D.N.Y. 2023, consolidated      in multidistrict litigation emphasizing mass reproduction as infringement)
• UMG Recordings v. Uncharted Labs (d/b/a Udio)      (S.D.N.Y. 2024, music industry suit over AI-generated audio trained on      copyrighted songs, with 2025 updates on damages).

Authorship and Ownership of AI Output

U.S. copyright law mandates human authorship for registration, creating ownership issues for AI-generated content lacking sufficient human creative input. Relevant references include:

• Thaler v. Perlmutter (D.D.C. 2023, aff'd D.C. Cir.      Mar. 2025, ruling that AI systems like the "Creativity Machine"      cannot be authors, as copyright requires human authorship)
• U.S. Copyright Office's Zarya of the Dawn decision      (2023, partial registration granted only for human-arranged elements,      excluding AI-generated images)
• U.S. Copyright Office's AI Report Part 2 (Jan. 2025,      affirming protection only for works with human-determined expressive      elements).

Patent Law Challenges

Patent systems worldwide require a human inventor, rejecting AI as the sole originator and leaving ambiguity on the level of human involvement needed for claims. Pivotal cases encompass:

• Thaler v. Vidal (E.D. Va. 2021, aff'd Fed. Cir. 2022,      holding AI like DABUS* cannot be an inventor under U.S. law)
• Thaler v. Comptroller-General of Patents, Designs and      Trade Marks (UK Sup. Ct. 2023, confirming inventors must be natural      persons)
• Commissioner of Patents v. Thaler (Fed. Ct. Aust.      2022, overturning prior ruling to deny AI inventorship)
• European Patent Office (EPO) J 8/20 (DABUS) (2021,      reiterated 2024 denial, emphasizing human inventors)
• Swiss Federal Administrative Court decision (Aug.      2025, allowing more flexibility for AI-assisted inventions but requiring      human naming).

Trade Secrets Protection

AI companies rely on trade secrets for model details, leading to litigation over misappropriation via employee mobility or reverse engineering. Notable cases include:

• Waymo LLC v. Uber Technologies, Inc. (N.D. Cal. 2017,      settled 2018, involving self-driving AI tech poached by former employees)
• OpenEvidence v. Pathway Medical (filed Feb. 2025,      alleging extraction of medical AI data as trade secret theft)
• Proper Fuels Inc. v. Phillips 66 Co. (Cal. state ct.      2024, $605M verdict for AI-related fuel optimization secrets)
• xAI trade secret dispute (2025, involving departed      employees leveraging proprietary AI features).

Footnotes:

Footnote 1: Fair use permits limited reproduction of copyrighted material without permission for purposes like criticism, commentary, news reporting, teaching, scholarship, or research, balancing public interest with creators' rights. Reference: 17 U.S.C. § 107.

Footnote 2: DABUS, an artificial intelligence system created by Dr. Stephen Thaler, was designated as the inventor in patent applications for inventions it allegedly generated autonomously, sparking global legal debates on AI inventorship. 

Courts in jurisdictions including the United States, United Kingdom, and Australia have consistently ruled that only natural persons qualify as inventors under existing patent laws, rejecting AI as a valid inventor. Reference: Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022)

Footnote 3: The quantum of human input refers to the minimum degree of human creative control, selection, and expressive contribution required for a work produced with artificial intelligence assistance to qualify as copyrightable under U.S. law. Courts have consistently held that purely AI-generated material lacks the human authorship demanded by the Constitution and the Copyright Act, leaving the precise threshold of meaningful human involvement as a key unresolved question in hybrid creations. Reference: Thaler v. Perlmutter, No. 22-cv-01564-BAH, 2023 WL 5333330 (D.D.C. Aug. 18, 2023), Affirmed No. 23-5233, 2025 WL 864244 (D.C. Cir. Mar. 18, 2025).