On September 1 the Justice Department filed 19 pages telling a New York judge that training an AI model on copyrighted writing is fair use. The DOJ OpenAI copyright brief is good news for model makers and close to irrelevant for the firm that buys their tools. That firm’s exposure sits in what it publishes and where it sells, and neither is settled by a statement of interest.
The filing landed in the consolidated New York case where the New York Times and other publishers are suing OpenAI and Microsoft, according to Bloomberg Law. The government isn’t a party. The publishers have pushed back, and Judge Sidney Stein has no obligation to agree with anyone. Our earlier look at AI-generated video and copyright risk drew the same line between how a model learns and what a business puts out.
Key points
- What does the DOJ OpenAI copyright brief actually say? It says training a model on copyrighted writing is fair use and asks the court to reject arguments otherwise.
- Does a training win cover what my business publishes? No. You didn’t train the model.
- What should a 50-person firm check this week? Four things, written down. Name the plan you’re on and what its indemnity covers.
- What does an edit trail look like in practice? Dull, and that’s the appeal. Keep the prompt, the raw draft and the edited version somewhere a colleague can find them, with a name and a date on the edit.
What does the DOJ OpenAI copyright brief actually say?
It says training a model on copyrighted writing is fair use and asks the court to reject arguments otherwise. It also says output questions should be judged separately and concedes that some output uses may not count as a new purpose. Nothing in it binds the judge, because a statement of interest only reports the government’s view.
Per a legal summary of the filing, the government argues training changes the works’ purpose so thoroughly that it’s fair use, and warns that heavy licensing requirements could entrench a few large technology firms. The same summary says the brief rejects a theory that AI output can harm a publisher’s market without copying its text. Gizmodo’s report adds that the brief wants output considered apart from training. That last point is the one a buyer should read twice.
A filing the judge can ignore still matters, because it tells vendors and plaintiffs where the government stands, and that shapes settlements and licensing talks long before any ruling. It doesn’t shape your liability. That’s the gap this column is about.
One caveat on our own work. We haven’t read the filing. This column rests on four secondary reports, and they differ on whether the brief makes a national security argument, so we don’t lean on it.
Does a training win cover what my business publishes?
No. You didn’t train the model. You publish what it writes, and you may sell into places where US fair use doesn’t apply. A court ruling on training leaves both of those exposures open, and most coverage skips straight past them. Your risk sits in outputs, contracts and the markets you sell into.
Take outputs first. The brief itself separates them from training and allows that reconstructed or redistributed material may not count as a new purpose. Picture a marketing lead pasting three paragraphs that closely track a publisher’s article onto your blog. A training ruling does nothing for that page.
Then jurisdiction. The Next Web notes that the EU has no fair use doctrine, only a closed list of exceptions, and that the AI Act’s Article 53 asks general-purpose model providers to respect rightsholders’ opt-outs. It also reports a Munich court ruling against OpenAI over memorized song lyrics, now under appeal. A US win wouldn’t travel. German courts are already testing AI liability, as our piece on the Munich ruling over AI disclaimers explains. Disclaimers fail in Europe for the same reason, and AI disclaimer liability now has a Munich ruling behind it.
What should a 50-person firm check this week?
Four things, written down. Name the plan you’re on and what its indemnity covers. Name who edits AI text before it goes out. Note if you sell into the EU. Note if any client contracts promise original work. None of it needs a lawyer to start. A spreadsheet and an afternoon will do.
Authorship is just as unsettled on the output side, where the Claude watermark leaves the rule to you.
The first check has the most moving parts. OpenAI’s Copyright Shield, announced in November 2023, reportedly covers ChatGPT Enterprise and API customers for output claims, not free or Plus users. Proskauer’s summary doesn’t spell out the carve-outs, so read your current terms and skip any 2023 summary, ours included. If your team writes client work on a personal plan, you may have no indemnity at all. Who owns the work you produce is a separate matter, which our piece on AI-generated work and copyright takes up. For comparing vendor terms side by side, start with our ChatGPT, Claude and Gemini business comparison.
What does an edit trail look like in practice?
Dull, and that’s the appeal. Keep the prompt, the raw draft and the edited version somewhere a colleague can find them, with a name and a date on the edit. Set a rule that long passages from a tool don’t go out untouched, and decide what “long” means for you. A paragraph is a reasonable line.
The trail does two jobs. If someone alleges a page copies their work, you can show a human read it and changed it. And if your vendor’s indemnity excludes material edits, you’ll know which edits you made. We’d put this on whoever signs off client deliverables, not on the person who wrote the prompt. The person who writes the prompt is rarely the person who gets the letter.
None of this costs money. It costs a habit, and habits are what small firms skip first.
Which contract clauses matter more than the ruling?
Three clauses matter more than the ruling: what the indemnity covers (outputs, inputs or both, and on which plans), the carve-outs that remove cover, and who controls the defence. An indemnity you can’t invoke without the vendor’s consent is a weaker promise than it reads, so check all three before you rely on it.
Carve-outs usually remove cover if you ignored a filter, fed in material you didn’t own, or edited the output in ways the vendor calls material. Here’s a worked example, and it’s ours, not a case. A 60-person agency posts an AI-drafted client article from a personal-tier account. A publisher alleges the piece reproduces its reporting. The agency’s contract promised the client original work. The model vendor’s indemnity, if one exists on that plan, is a separate contract the client never signed. The agency holds the risk. Each link in that chain can be checked against your own paperwork, which is the point.
Reading the paper matters as much as watching the court. Our look at the Musk v. OpenAI verdict and vendor contracts makes the same case from a different dispute. Timing decided another big case, and the Musk OpenAI verdict turned on a calendar.
A fair counter says we’re overcomplicating this. If the government and probably the courts treat training as fair use, AI gets cheaper, vendors defend their customers, and small firms should get on with it. The brief does tilt the odds toward model makers on training. But “training is probably fine” and “we’re covered” are different claims, and the second depends on a contract most owners haven’t opened since they clicked accept. We’d change our view if Judge Stein ruled on outputs as well as training and sided with the vendors on both, or if the big vendors extended their indemnities to every paid tier without carve-outs. Nothing here is legal advice, and we can’t say how the judge will rule.
Reply briefs on the summary judgment motions are due November 6, 2026, per the legal summary. Open your plan’s terms before then.
Frequently asked questions
Does the DOJ brief make AI training legal?
No. A statement of interest gives the court the government’s view and doesn’t bind it. Judge Sidney Stein will decide the fair use question, and the ruling could be appealed.
Am I covered if my team publishes AI-written content?
It depends on your plan and your vendor’s terms. OpenAI’s Copyright Shield reportedly covers ChatGPT Enterprise and API customers for output claims, but not free or Plus users. Read the carve-outs in your current agreement.
Does a US fair use ruling apply in Europe?
No, not directly. The EU has no fair use doctrine and relies on a closed list of exceptions, so a US win on training would not carry over to sales or publishing there.
Archive entry dated September 7, 2026, written and fact-checked on October 8, 2026. Sources are linked on the claims they support.
