OwnershIP Disputes For Neural-Model-Produced Urban Traffic-Decongestion Algorithms

18 Mar 2026 --
0 Comments

📌 Core Legal Issue: Who “owns” a neural model’s output?

When an AI or neural network produces an algorithm (e.g., a traffic‑decongestion algorithm), the legal dispute hinges on whether the output is eligible for intellectual property (IP) protection and who the law recognizes as the owner. Traditional IP regimes presume human authorship or inventorship. If none exists, courts must decide:

Is the AI a “creator/inventor” (usually not under current law)?
Does ownership belong to the human who trained or used the AI?
Does corporate/contractual control decide ownership?
Can there be joint ownership between humans and machines?

Below are five detailed legal disputes that illuminate how these issues are handled.

1) GEMA v. OpenAI (Regional Court of Munich, 2025) — Copyright Infringement and Memorisation

Facts:
GEMA, the German music rights society, sued OpenAI alleging that its language models were trained on copyrighted song lyrics without permission. GEMA claimed the AI memorized and reproduced protected lyrics, violating exclusive rights under German/EU law.

Key legal questions:

Can a neural model’s training process infringe copyright?
Does the model “reproduce” protected works if its outputs contain verbatim or substantially similar content?

Court’s Holding:
The Munich court held for GEMA: generative AI infringes when it reproduces copyrighted material that is memorized in a way that acts like reproduction under copyright law. The decision treats unauthorized training as an infringement even if the output is created by algorithms and not by human copying.

Why this matters:

It signals that AI training without licensing can violate creators’ rights, giving authors a basis to claim ownership interests or seek damages when output resembles their works.
Although not directly an ownership dispute over an algorithm’s ownership, it affects legal ownership claims because if an output incorporates protected material, the rights of original creators remain paramount.

2) Thaler v. Perlmutter / U.S. Copyright Office (AI‑Generated Art) — Authorship & Human Requirement

Facts:
Dr. Stephen Thaler sought copyright for an image that his AI system “DABUS” generated autonomously, listing the AI itself as the author and Thaler as owner. The U.S. Copyright Office rejected it because copyright requires a human author. Lower courts upheld that rejection. The U.S. Supreme Court later declined to hear the appeal.

Legal principles:

U.S. law requires human authorship for copyright protection.
An AI, being non‑human, cannot hold or transfer copyright.

Judicial Reasoning:
The courts affirmed that because the work was generated by a machine without sufficient human creative contribution, it lacked the necessary human authorship under the Copyright Act.

Implications for AI‑Created Algorithms:

If an algorithm is fully generated by an AI with no meaningful human input, it may not be eligible for copyright, leaving it in the public domain.
Where a developer or user significantly directs the model’s creation (e.g., designing the architecture, tuning parameters), courts may find enough human intellectual input to allocate ownership to that person or entity.

3) DABUS Patent Disputes (Multiple Jurisdictions) — Inventorship & Ownership

While not exactly the same as traffic algorithms, the DABUS cases on AI‑generated inventions are core to understanding ownership of AI outputs:

Facts:
Patent applications listing an AI (“DABUS”) as the sole inventor were rejected by patent offices in the U.S., EU, and UK because an inventor must be a human. South Africa uniquely granted a patent listing DABUS as inventor, but most jurisdictions rejected this.

Ownership outcome:

Where patents were rejected, ownership rights could not be granted because there was no recognized human inventor.
Where inventions are made using AI, the human who provided the creative inventive step is usually deemed the inventor.

Relevance to neural models:
For algorithms developed by AI, if the law refuses to recognize the model as inventor, ownership defaults:

to the person who guided the AI, or
to the employer under “work made for hire”, or
as determined by contract/licence.

4) AI Training and Reproduction Disputes (U.S. & EU) — Fair Use and Rights

While not directly about ownership of outputs, ownership disputes often arise when outputs resemble copyrighted material:

U.S. courts have sometimes found AI training to be fair use in copyright suits, allowing developers to use copyrighted works in training, but liability still may arise depending on how outputs replicate content.
Others (e.g., major author lawsuits) allege copyright violations for training on copyrighted books, affecting how outputs are treated and whether creators retain ownership interests.

Connection to ownership:
If outputs infringe other creators’ rights, the courts may deem them unattributable or subject to compensation rules, complicating who has lawful ownership or exploitation rights.

5) Comparative Case: Nova v. Mazooma Games Ltd. (UK, CDPA Section 9(3)) — Computer‑Generated Works

This is a classic precursor case (in the UK) applying a statutory rule that computer‑generated works can have a human author defined as “the person by whom the arrangements necessary for creation were undertaken.”

Fact pattern:
Though not AI specifically, the UK law (and UK courts applying it) have recognized that where a computer creates a work, the person who organized the necessary steps can own the copyright.

Takeaway:

Ownership can attach to a human who structured and directed the AI’s activity, even if the machine executed the work.
This principle could analogously apply to AI‑generated algorithms: the human who devised the neural network architecture, training data, and parameter choices arguably owns the output.

⚖️ Key Legal Themes Across These Cases

🧠 1. Human Authorship/Inventorship is Central

Current IP systems usually require a natural person to qualify for copyright or patent rights. Courts repeatedly hold that an AI itself cannot be an owner because it is not a legal person.

🤖 2. Amount of Human Involvement Matters

Where a human sets parameters, curates outcomes, or infuses creative judgment, courts are more likely to assign ownership to that person.

📄 3. Contract ≠ Common Law

Many ownership disputes are solved by contract (licenses, terms of use, employment agreements). Without clear contractual provisions, default legal standards apply.

📍 4. Jurisdictional Variance

Different countries treat AI outputs differently:

U.S.: Strict human authorship/inventorship requirement.
Europe: Infringement and reproduction rules may still apply; some national laws allow assigned authorship for computer‑generated works.
China: Some courts are more flexible in recognizing human guided AI output as protectable when there is intellectual input.

📊 5. Outputs That Replicate Copyrighted Input Create Liability

If training data or outputs amount to reproduction of existing copyrighted works, ownership disputes are further complicated by infringement claims.

📌 How This Applies to Urban Traffic‑Decongestion Algorithms

If a neural model produces an algorithm:

If human engineers designed the AI’s architecture and iterative tuning, they (or their employer) will likely be recognized as owners.
If the output replicates proprietary code/data of others, original rights holders may have claims or co‑ownership interests.
If the AI acted entirely autonomously, some jurisdictions may categorize the algorithm as uncopyrightable and unpatentable, leaving it unowned by IP law.

Summary of Case Law Covered

Case/Dispute	Jurisdiction	Legal Focus	Key Ownership Outcome
GEMA v. OpenAI	Germany/EU	Copyright & AI training	AI can infringe by reproducing copyrighted works
Thaler v. Perlmutter	U.S.	Authorship requirement	No copyright without human author
DABUS Patent Cases	Multiple	Inventorship	AI cannot be inventor; humans must be
Anthropic Author Suits / Fair Use	U.S.	Training data copyright	Fair use defense affects output rights
Nova v Mazooma	UK	Computer‑generated works	Human who arranged creation can be author