You Don’t Own the Code That AI Writes for You: A Problem of Ownership in the Age of Generative Coding

J. Rogers, SE Ohio

Abstract

The rapid adoption of generative AI coding tools has created a quiet legal crisis. Companies are replacing human developers with AI, building massive codebases from machine‑generated output, and assuming that they own the resulting intellectual property. Under current US copyright law, this assumption is largely false. This paper explains why purely AI‑generated code cannot be copyrighted, why the distinction between “AI‑generated” and “AI‑assisted” matters, and how the rush to replace human programmers is producing millions of lines of legally orphaned code. The paper concludes with practical risks and recommendations for organizations relying on AI coding tools.

1. Introduction

In 2026, a software engineer in Stockholm told the New York Times: “I probably spend more than my salary on Claude.” Across the industry, AI coding assistants are being framed as efficiency miracles. Yet a fundamental legal question has been largely ignored: Who owns the code that an AI writes for you?

The answer from the US Copyright Office, federal courts, and the Supreme Court (by refusal to hear the contrary) is clear: No human author, no copyright. If a company’s codebase contains large volumes of purely AI‑generated code, that code is effectively public domain. Competitors can copy it. Security flaws can be copied without remedy. And the company cannot sue for infringement.

This paper outlines the legal landscape, the critical distinction between “generated” and “assisted” works, and the real‑world consequences for businesses that are currently firing junior developers and replacing them with AI.

2. The Legal Foundation: Human Authorship Is Required

2.1 The Copyright Act of 1976

US copyright law protects “original works of authorship fixed in any tangible medium of expression.” The Supreme Court has repeatedly held that an “author” must be a human being. In Burrow‑Giles Lithographic Co. v. Sarony (1884), the Court defined author as “he to whom anything owes its origin.”

2.2 The Copyright Office’s Clear Position

In 2023, the US Copyright Office issued policy guidance stating that it “will register an original work of authorship, provided that the work was created by a human being.” Works generated by artificial intelligence with no human creative contribution will not be registered.

Applying this to code: If you ask an AI “Write me a function to validate an email address,” and you copy‑paste the output without meaningful human modification, that function is not copyrightable. Anyone may legally copy it.

2.3 Thaler v. Perlmutter – The Supreme Court Refuses to Intervene

Dr. Stephen Thaler attempted to register a work he said was created entirely by an AI system. The Copyright Office refused. The district court and the DC Circuit affirmed. In March 2026, the Supreme Court declined to hear the appeal, letting the lower rulings stand.

The DC Circuit’s opinion is blunt: “Copyright law requires an ‘author’ – a human being.” The court noted that the Copyright Act uses terms like “children,” “grandchildren,” and “widow,” which only apply to natural persons. The outcome is settled: purely AI‑generated works have no copyright protection.

3. The “Generated” vs. “Assisted” Distinction

The Copyright Office draws a critical line:

AI‑Generated	AI‑Assisted
The AI produces the expression with no human creative control over the specific form.	A human uses AI as a tool, but provides sufficient creative input – editing, selecting, arranging, rewriting.
Not copyrightable.	May be copyrightable (the human’s contributions are protected).

In practice, most current use of AI coding tools leans heavily toward “generated.” Engineers type a prompt, receive code, and commit it without meaningful change. This is exactly the scenario the Copyright Office describes as lacking human authorship.

The Office has explicitly warned that iterative prompting – asking the AI to refine its output – does not automatically confer authorship. Unless the human makes original, creative modifications to the AI’s output, the result remains unprotectable.

4. Why “Sweat of the Brow” Doesn’t Matter

Some argue that because they spent hours crafting prompts, they “worked hard” and should own the result. Copyright law rejected “sweat of the brow” decades ago. In Feist Publications, Inc. v. Rural Telephone Service Co. (1991), the Supreme Court held that effort alone does not create copyright; there must be original creative expression.

A perfect prompt that generates perfect code still produces a work whose expression originates from the AI, not the human. Unless the human alters that expression creatively, there is no copyright.

5. Real‑World Consequences for Companies

5.1 Your Codebase Is a Legal Orphan

If a competitor copies a purely AI‑generated function from your product, you have no copyright infringement claim. The code is not yours in the eyes of the law. This undercuts the entire value proposition of proprietary software.

5.2 M&A Due Diligence Nightmare

When a company is acquired, the buyer’s lawyers will ask: “What percentage of your code was AI‑generated without human modification?” A high percentage could make the target’s “crown jewel” IP worthless. Deals will collapse or valuations will crater.

5.3 Security and Liability Traps

You cannot retroactively copyright AI‑generated code that already exists. If that code contains a vulnerability, and a competitor copies it, you have no legal recourse. Worse, if the AI reproduced code from its training set that is subject to a restrictive license (GPL, etc.), you could be sued for infringement by the original human author – while you still own none of your own output.

5.4 The “Vibe Coding” Fad Is Self‑Defeating

The current trend of “vibe coding” – describing an app idea to an AI and committing whatever it produces – is legally catastrophic. Entire startups are being built on code that belongs to no one. Investors who discover this will walk away.

6. Can You Protect AI‑Generated Code Any Other Way?

Copyright is not the only form of IP, but the alternatives are weak:

• Trade secret – Protects confidential information, but offers no protection if the code is reverse‑engineered or independently discovered. Once AI‑generated code is distributed (e.g., in a compiled app), trade secret protection is largely lost.
• Patent – Might cover algorithms, but most routine code does not meet the novelty and non‑obviousness requirements. AI‑generated code is unlikely to be patentable.
• Contract – Terms of service can restrict users, but contracts don’t bind competitors who never agreed to them.

In short, there is no substitute for copyright for protecting the literal expression of software code.

7. Recommendations

For organizations using AI coding tools:

1. Audit your codebase – Identify which files or functions were AI‑generated with minimal human modification. Flag them as unprotectable.
2. Change workflows – Require that a human engineer meaningfully edit, rewrite, or arrange any AI‑generated output before committing. Document the creative changes.
3. Maintain a “human authorship” log – Record who modified what, and what creative choices were made.
4. Do not replace junior developers – Juniors are the humans who will provide the creative modifications needed for copyright. Without them, you are building an unownable codebase.
5. Consult legal counsel – The law is evolving. The EU and other jurisdictions may take different approaches. But under current US law, the risk is real and severe.

8. Conclusion

The narrative that AI coding tools are simply “efficiency” ignores a foundational legal reality: you do not own what you do not create. When companies fire entry‑level engineers and replace them with AI, they are not just losing future senior talent – they are losing the legal ability to claim ownership over their own product.

The seed corn is being sold. The code being written today may be free for anyone to take tomorrow. And the executives celebrating quarterly stock bumps will be long gone when the lawyers arrive to ask: “Who wrote this – and do you have the papers to prove it?”

The answer, more often than not, will be no.