The DOJ told a judge to rule for OpenAI in the New York Times case. Your archive's price is on the line.

What happened

The Washington Post reported on Sept. 2, 2026 that the U.S. Department of Justice filed in The New York Times Co. v. Microsoft Corp. and OpenAI and urged the judge to rule for OpenAI and Microsoft. Axios followed on Sept. 8 with coverage of the case heating up. Court filings reported by GV Wire on Sept. 5 show both sides arguing fair-use doctrine.

The Times sued in December 2023. In July 2026, it accused OpenAI of withholding evidence in discovery, per TechCrunch. The case is now at summary judgment on fair use — the stage where a judge can end it or send it to trial.

Two disclosures. I have the Post's report at headline level and have not verified the docket number or the legal form of the DOJ filing — statement of interest or amicus brief. And no ruling has been issued. If you see a post this week saying the court decided, it is wrong.

Why the DOJ part is the actual news

A private copyright suit between a newspaper and two tech companies does not normally attract the executive branch. This one did.

That filing is not binding on the judge. It is a signal about how the current administration reads fair use for AI training, and that signal travels. There are pending suits against Anthropic, Meta, Perplexity and Brave working the same ground. One filing, many defendants.

The question in front of this judge is whether your archive is raw material that anyone can take for free, and the U.S. government has asked him to say yes.

What it actually changes for you

Every AI licensing deal signed since 2024 was priced against litigation risk. That is the only reason the numbers were non-zero. If training on copyrighted work is fair use, the risk premium disappears and renewals get renegotiated down — possibly to nothing. If the Times wins, every model trained on unlicensed news becomes a liability and deal values move the other way.

That applies to you even if you have never licensed anything. Your rate for a sponsored series, your leverage in a syndication deal, the value of a back catalog you might sell one day — all of it sits on top of the same question.

The distinction almost everyone will miss

Training and output are two separate legal questions.

"Can a company ingest my archive to train a model?" is not the same as "can it publish a summary that reproduces my reporting?" A March 2026 ruling covered by Copyright Lately held that AI news summaries may infringe. That is the doctrine News Corp is using against Brave, in a suit filed this summer.

So a publisher can lose on training and still win on summaries. If the headline next month reads "court rules AI training is fair use," that does not mean AI products can republish your work — and confusing the two will cost you money in your next negotiation.

There is also a quieter third track. Reddit's suit against Perplexity over data scraping mostly survived a motion to dismiss on July 31, per Law360 and Reuters. That case runs on contract and terms-of-service theories, not copyright. It does not depend on fair use at all. For small publishers, that route may matter more than this one.

What this story is not

It is not a decision. It is not a settlement. There is no confirmed damages figure anywhere in this coverage, and anyone quoting one is guessing. The DOJ does not decide the case.

What to do about it

1. If you have an AI licensing deal, find the renegotiation clause today. Ask your lawyer whether a fair-use ruling triggers a reprice. Nobody has reported how common that clause is.
2. Put your terms of service in order. The Reddit track rewards publishers whose contracts say plainly what bots may not do.
3. Keep dated records of what you published and when. In discovery, provenance is the asset.
4. Do not sign a multi-year archive license this quarter at pre-ruling prices without a review clause.
5. When the ruling lands, read which question it answered — training or output — before you tell your audience what it means.
6. Watch the Brave and Perplexity cases in parallel. They are where the summaries question gets settled, and that is the one that touches your traffic.