Protection Of Neural Architecture Search Algorithms Generating Self-Optimizing Ai Systems.

16 Feb 2026 --
0 Comments

1. Understanding the Context

Neural Architecture Search (NAS) algorithms:

Automatically design neural network architectures to optimize performance on specific tasks.

Generate self-optimizing AI systems without human intervention in the architecture design.

Can produce outputs ranging from machine learning models to AI-driven applications.

Legal Issues:

Copyrightability of the NAS algorithm itself (software copyright, code protection).

Copyright of the AI systems generated by NAS. Are these derivative works or entirely new works?

Authorship questions: Who owns the copyright of self-optimizing AI systems—developer, user, or the AI itself?

Patent vs. copyright: NAS-generated AI systems may also raise patentability questions.

2. Relevant Legal Principles

a. Software Copyright

Source code and object code are copyrightable as literary works in most jurisdictions.

b. AI-Generated Works

U.S. and EU law: Only humans can hold copyright. Autonomous AI cannot be recognized as an author.

Human intervention or guidance is required to claim copyright.

c. Derivative and Transformative Works

NAS outputs may be derivative of existing code or AI frameworks.

Protection depends on originality and human creative input.

3. Key Case Laws

Here are seven relevant cases analyzed in detail:

1️⃣ Thaler v. U.S. Copyright Office (2022)

Facts: Stephen Thaler sought copyright for artwork produced entirely by AI (“DABUS”).

Holding: Courts rejected AI authorship; only humans can be authors.

Implication for NAS: Self-optimizing AI generated autonomously by NAS cannot hold copyright. Human developers who design, guide, or modify NAS outputs may claim authorship if they exert creative control.

2️⃣ Naruto v. Slater (2018)

Facts: Monkey took selfies using a photographer’s camera.

Holding: Non-human entities cannot hold copyright.

Implication: Reinforces that NAS-generated systems, like autonomous AI or animals, cannot independently claim copyright.

3️⃣ Feist Publications v. Rural Telephone Service (1991)

Facts: Telephone directory copied by Feist Publications.

Holding: Facts themselves are not copyrightable; only original expression is.

Implication: NAS outputs using standard algorithms or publicly available components may lack originality unless the developer adds creative expression or custom design.

4️⃣ Oracle v. Google (2016)

Facts: Google used parts of Java API in Android OS.

Holding: Use of APIs may qualify as fair use; copyright protects expression, not ideas.

Implication: NAS algorithms often build on existing frameworks (TensorFlow, PyTorch). The expression of original code can be protected, but general algorithmic ideas are not.

5️⃣ GitHub Copilot Licensing Cases (Ongoing)

Facts: Copilot AI trained on open-source code raises questions of copyright infringement.

Implication: NAS-generated code may inadvertently reproduce copyrighted code. Proper licensing compliance is necessary to avoid infringement.

6️⃣ Nova Productions v. Mazooma Games (2007, UK)

Facts: Video game used copyrighted film footage.

Holding: Creative arrangement of content can confer copyright.

Implication: NAS outputs curated or modified by humans—for example, human-devised objective functions or network constraints—can have copyright protection.

7️⃣ Feit v. Sony (Software Copyright, 2000s)

Facts: Software with autonomous functions challenged for originality.

Holding: Courts emphasized human authorship and originality in software expression, not automated functions.

Implication: NAS-generated self-optimizing AI systems cannot claim copyright as standalone creations, but the human-designed NAS algorithm and any human creative choices can be protected.

4. Practical Application

Scenario	Copyright Protection	Notes
NAS algorithm code	✅ Protectable	Software copyright protects the written code and implementation
AI systems generated fully autonomously by NAS	❌ Not protectable	No human authorship
NAS outputs curated/modified by humans	✅ Protectable	Human intervention adds originality, allowing copyright
Use of existing frameworks (TensorFlow, PyTorch)	⚠️ Conditional	Only original expression of implementation protected; idea is free
AI-generated architecture based on public domain data	✅ Protectable if human-guided	Human-defined objectives and design choices confer copyright

5. Key Takeaways

Human authorship remains central: Autonomous NAS outputs alone cannot be copyrighted.

NAS algorithm code itself is protectable: Software copyright covers implementation, documentation, and source code.

Human-guided outputs are eligible for copyright: Developers who intervene, modify, or curate NAS outputs can claim protection.

Derivative works require careful licensing: NAS often builds on open-source or third-party frameworks; compliance is critical.

Court precedent emphasizes originality over automation: Thaler, Naruto, Feist, Oracle v. Google collectively stress creative human contribution.