AI Makes Securing Copyright Protection for Software Code Tricky Guide

Copyright protection for software code is being sacrificed, knowingly or not, for the speed and efficiency of AI coding.

This rapid shift in the role of humans from writing code to managing artificial intelligence tools upends traditional copyright protection strategies. Original human-written code is generally copyrightable. But AI-generated code that lacks human authorship is ineligible for copyright protection under US law.

“Vibe coding”—where humans describe a desired software program in natural language and GenAI tools write the code—is pervasive. This isn’t limited to the tech industry. Employees across industries are vibe coding software solutions, which can be valuable to employers.

Developers estimate 42% of code is AI-generated or assisted and the number was expected to increase significantly, according to an October 2025 survey.

The lack of copyright protections is a big deal. While other legal protections for software may be available, they have limitations. Trade secrets may be reverse engineered and public disclosure may destroy protection.

Utility patents may protect novel processes or methodologies, but are subject to strict eligibility requirements and public disclosure. Contracts require proper formation and may be subject to capped damages or other limitations.

It seems unwise to abandon copyright, a proven legal regime providing advantages such as statutory damages and attorneys’ fees.

Moving Fast

Things are moving fast. Agentic coding tools such as Anthropic’s Claude Code and OpenAI’s Codex enable humans to direct agents in multi‑step planning and execution on complex coding tasks. Viral general-purpose agent platforms such as OpenClaw recently turbocharged the excitement surrounding agent capabilities.

Rather than abandon copyright protection strategy, refocus it. Start with these principles:

• Copyright generally doesn’t protect purely AI-generated material.
• Copyright can protect original human-created expression within a work even if the work also includes some AI-generated material.
• Whether human contributions are sufficient to constitute copyright authorship must be analyzed on a case-by-case basis.
• Human-written prompts alone are generally insufficient for copyrightability of AI-generated outputs.

The key is documenting creative human contributions throughout the software development lifecycle—not just coding. This isn’t straightforward in agentic workflows because the boundary between human and agent behavior is often blurred, making it difficult to separate human-driven creativity from machine‑generated code.

The ability to secure copyright protection can be aided using the following copyright protection stack:

Human-written design and architectural specifications: Non-code artifacts may be the IP protection gold. Even if AI agents write the code, humans provide specifications to guide them, and those specifications may be copyrightable.

This includes prompts and prompt scaffolding, architectural and design plans, implementation frameworks, orchestration code and protocols, contextual materials, curated datasets, product requirements, guardrails and governance policies, data flow diagrams, flow charts and pseudocode.

The most sensitive of these artifacts may be so valuable as to justify protecting them only as trade secrets rather than risking the limited public disclosure required for copyright registration. This is a strategic call.

Human-written code identified separately from AI-generated code: It’s unclear how best to document human contributions in agentic coding workflows. Is manual tracking realistic such as, human-written notes, logs, or commit annotation and tags to record human contributions?

Documenting rejection of code suggestions may be equally important in showing human creative direction and control. Some tools even claim to track line-by-line human versus AI code contributions. But the effectiveness of automated attribution tools as evidence of copyrightability is untested. Whether human contributions in agentic workflows can be reliably documented could be a trillion-dollar question.

Substantial human modifications to AI-written code: Humans may modify AI-generated code to such a degree that the modifications are copyrightable. The Copyright Office pointed to GenAI tools that allow human users to make substantial edits to AI-generated text, highlight regions for the model to focus on, use built-in tools to request, accept, and reject in-line suggestions, and write instructions that detail particular edits to be made.

Whether such modifications rise to the minimum level of originality requires a case-by-case determination. If they do, the human contributions should be copyrightable. Contemporaneous documentation is key.

Human selection, coordination, and arrangement of AI-generated code: Even if a human doesn’t write a single line of code, copyright protection may be possible for creative selection, coordination, or arrangement of AI-generated code. For example, human selection of numerous AI-generated code blocks, plus creative ordering, grouping, and interconnection of those blocks, may suffice.

Merely selecting and arranging two or three AI-generated code outputs may be insufficient creative choice. Electing obvious or standard components, or arranging them in a standard or obvious architecture, may also fall short. The key is bespoke curation into a creative whole from many options.

Such “compilation” protection is considered “thin” or weak because it protects the overall structure rather than the code itself. But it could be valuable where the secret sauce is the architecture rather than the code—perhaps precisely where agentic coding is heading.

Intentionally building these layers into coding workflows strengthens the ability to copyright.

The Copyright Office says it has registered hundreds of works that incorporate AI-generated material, with the registration covering the human author’s contribution to the work.

Deceptively Easy

Applying for copyright registration of AI-assisted code is deceptively easy. Applicants can simply disclaim “source code generated by artificial intelligence” as excluded material, without providing line-by-line attribution. But obtaining a registration certificate from the Copyright Office isn’t the same as successfully enforcing copyright in court.

It’s unclear how copyright registrants can best prove copying of creative human contributions within agentic workflows in infringement litigation.

Copyright remains a central pillar of legal protection for software and demands careful planning. Build the copyright protection stack into coding workflows, including contemporaneous documentation of human contributions through manual and automated processes.

Remember that code isn’t the only asset. Architectural and orchestration artifacts may be the new moat that drives protectability and competitive advantage in the agentic era.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law, Bloomberg Tax, and Bloomberg Government, or its owners.

Author Information

Michael Justus is a shareholder at Carlton Fields with more than 15 years of litigation experience in intellectual property, advertising law, and emerging issues involving artificial intelligence and technology.

Write for Us: Author Guidelines