This is AI Law, on Terms.Law Radio. One oh one point three on your imaginary dial. The station where I read the terms of the machines. Tonight's question comes up in almost every consult I do now: who owns what your AI makes, and who pays when it gets something wrong?

Start with ownership, because most people have it backwards. In the United States, copyright protects human authorship. The Copyright Office has said, repeatedly and in writing, that material generated entirely by a machine, with no human creative control, is not protected by copyright at all. Not owned by the AI company. Not owned by you. Not owned by anyone. Public domain. Now, that is the pure case. The moment a human meaningfully selects, arranges, edits, and directs the output, protection can attach to the human contribution. Which means the practical question for your business is not can I use this, it is can I stop anyone else from using it. If your product's core asset is raw AI output, your moat may be thinner than your pitch deck says.

Second: the license question hiding in your vendor's terms. When you generate something with a commercial AI service, two documents decide what you can do with it: the provider's terms of use, and increasingly, a separate business agreement if you are on a paid tier. Most major providers now say you own the output, or that they assign it to you. Read the next sentence too, the one nobody reads. It usually says the same output may be generated for other customers, and no exclusivity is promised. You own your copy. You do not own the idea of it.

Third: who pays when the AI is wrong. Suppose the model produces text that infringes someone's copyright, or fabricates a fact that costs your customer money, and your company shipped it. The claim lands on you, not on the model vendor. That is what the indemnification clauses are for, and this is where the fine print earns its reputation. Several large AI vendors now advertise copyright indemnities for business customers. The coverage is real, but it is conditional. Typically you lose it if you turned off the safety filters, if you knew the output was infringing, if you modified the output, or if you used a preview feature instead of the paid production tier. An indemnity you qualify for by accident is not a plan. If AI output goes into your product, someone at your company should be able to say, out loud, which conditions you are relying on and how you stay inside them.

Fourth: your own customers. If you build AI features into your service, your terms of service need to answer three questions your users will eventually ask. One: do you train on their data, and can they opt out? After this summer's wave of policy updates, users are primed to look for that clause, and the companies that hid it are wearing the bad grades on my Watchdog station to prove it. Two: who owns what your users generate inside your product? Three: what do you disclaim when the AI is wrong, and is that disclaimer actually enforceable in the states where your users live? Boilerplate copied from a bigger company's terms answers none of these, because their risk is not your risk.

Last one, quickly: the rules themselves are moving. Washington is pushing voluntary frameworks and challenging state AI laws, the states keep passing them anyway, and Europe's AI Act is phasing in on its own schedule with its own reach. I covered the federal side in the Washington to Contract report, on the Counsel's Desk, ninety two point five on this same dial. The practical takeaway for a small company is unglamorous: know which states your users are in, assume the strictest one sets your floor, and put one person in charge of noticing when the floor moves.

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 is moving fast; verify the current state of any rule before you act. My written analysis, the AI clause library, and the tools I build are at terms dot law. 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. Read the terms of your machines. Good night.