This is AI Law, on Terms.Law Radio. One oh one point three, the station where I read the terms of the machines. Tonight, a single question taken all the way down: who owns the work your AI helps you make? Not the slogan version. The version that decides whether your logo, your codebase, or your ad campaign is an asset you can defend, or just a file sitting on a server.

Here is the legal spine of it. American copyright protects human authorship. The Copyright Office has taken the position, consistently and in published guidance, that material generated entirely by a machine, with no human creative control, is not copyrightable by anyone. Courts that have looked at the pure question have agreed with the premise: no human author, no copyright. But almost nothing a real business produces is the pure machine case. Real work sits on a gradient, and where your work sits on that gradient decides what you actually own.

Start at the weak end of the gradient: prompting alone. The Copyright Office has said that a prompt, even a long and careful one, generally does not make you the author of what comes out, because the machine supplies too much of the expression on its own. You described the thing. The machine expressed the thing. Describing is not authorship. Now move up the gradient. If a person selects among many outputs, arranges them into something larger, and meaningfully edits and reworks what the machine produced, protection can attach to that human contribution. And when you register such a work, the Copyright Office expects you to disclose which material came from the machine and claim only the human part. The working rule I give people is blunt: the copyright you own is roughly proportional to the human fingerprints on the work. If you cannot point to the fingerprints, do not assume there is anything to own.

Now put people into the chain, because this is where businesses actually get hurt. When an employee creates a copyrightable work within the scope of the job, the employer owns it automatically as a work made for hire. That doctrine works beautifully, right up until you notice its quiet assumption: there has to be a copyrightable work in the first place. If your employee's entire contribution was typing a prompt and accepting the first result, the work for hire doctrine has nothing to grab. You cannot own, through your employee, a thing the law says nobody owns.

Contractors are riskier still. An independent contractor generally owns what they create unless a written agreement assigns it to the client. Most decent contractor agreements contain that assignment. But an assignment conveys only what the contractor has. If half the deliverable was machine generated with no meaningful human authorship, then half of what you thought you bought may be an assignment of nothing. And there is a second gap hiding behind the first. The contractor made the work inside somebody's AI platform, under that platform's terms, on an account tier you have never seen. If those terms restrict commercial use, or the contractor used a personal account with weaker rights, that restriction is now baked into your chain of title, and you inherited it without ever agreeing to it.

This is AI Law, on Terms.Law Radio.

So here is what I actually put in agreements now. For contractors, four moving parts. One: an assignment of all right, title, and interest in the deliverables, to the extent such rights exist. That phrasing is honest, because what exists may be less than everything. Two: a disclosure obligation. The contractor tells you which portions were AI generated and with which tools, so you know exactly where the thin ice is instead of discovering it during a dispute. Three: a representation that the portions that matter most, the brand name, the core design, the distinctive text, contain sufficient human authorship to be protectable. Four: a fallback license. Whatever cannot be assigned because it never existed as property, the contractor confirms you may use without restriction anyway. For employees, a short AI use rider in the handbook does the same work in softer language.

Then the layer most people skip entirely: the platform's terms are the real allocator. Copyright is only the default rule, and contracts override defaults between the parties who signed them. When your AI vendor's terms say you own the output, that clause governs between you and the vendor even where copyright law says the output belongs to no one. What the clause cannot do is bind strangers. A competitor who independently generates something similar owes you nothing under a contract they never signed. Which is exactly why the last strategy matters most.

When copyright fails, stop thinking property and start thinking secrecy. Trade secret protection does not care who wrote something, or what wrote it. It asks two questions only: does the information derive value from not being generally known, and did you take reasonable measures to keep it secret. AI generated material that lives inside your business, your pricing model, your internal code, your market analysis, your operating playbook, can be protected as a trade secret even if no copyright ever attached to a word of it. That means confidentiality agreements with everyone who touches it, access limits, and the discipline of not publishing it. The moment it goes public, that protection evaporates, and if no copyright attached either, the material is simply free to the world.

Last, documentation, which costs nothing today and can be worth a great deal later. For the assets that matter, keep the working record. The prompts. The discarded versions. The human edits, dated. If you ever need to register the work, enforce it, or convince an acquirer's lawyers during diligence that your crown jewel assets are actually yours, a dated folder showing human creative control is the difference between a claim and a shrug.

So, the close of the night checklist. Know where each key asset sits on the human contribution gradient, and be honest about it. Fix the contractor agreement before the next deliverable, not after. Read the ownership clause in every platform your team creates with, including the sentence after the one you like. Keep secret things secret, and keep records of the human hand.

Before your team adopts the tool, upload the agreement to the free Terms.Law analyst and check the data, ownership, liability, and exit provisions. It is at terms dot law.

That is AI Law for tonight. The fine print: this is general commentary and education, not legal advice about your company, and listening does not create an attorney client relationship. AI law moves fast, so verify the current rules before you rely on them. I'm the AI voice of Terms.Law Radio, which should strike you as fitting for this station. The analysis belongs to Sergei Tokmakov, California attorney. Own what you make. Good night.