The journalism lawsuits are not really about journalism

Another group of newspapers has sued OpenAI and Microsoft for mass copyright infringement, joining a queue that now includes The New York Times, eight Alden Global Capital titles in the United States, and a Canadian consortium asking for fifty billion dollars. The legal question is whether training large language models on copyrighted articles is fair use. The more honest question is who ends up paying for the journalism that taught these systems how to write, and who ends up owning what that journalism becomes. Those are different questions, and the second one is the one the lawsuits will not really settle.

The framing on offer. The press release version of this story is that the news industry is fighting back. Newsrooms produced the high-quality text that grounded the models; AI companies took it without asking; the courts will now decide whether they have to pay. Win the suit, save the journalism. It is a clean story and most coverage has run with it.

I do not think that is what is actually going on here. The lawsuits are real, the infringement claims are serious, and OpenAI's fair use defence is going to be tested properly for the first time. But the framing collapses three different fights into one, and the people who lose in two of those fights are not the same people who win in the third.

Who is actually at the table. OpenAI has already signed licensing deals with News Corp (reported at roughly 250 million dollars over five years), Axel Springer (reported in the multi-millions per year), and the Associated Press. These are the largest global news operators, the ones with in-house legal teams, the ones whose mastheads carry enough brand weight that a chatbot citing them is worth something to the chatbot.

The plaintiff list tells you who could not get to a table. Alden Global Capital's regional papers — the Chicago Tribune, the Orlando Sentinel, the New York Daily News. Canadian outlets pooling claims to reach a number large enough to matter. Local and regional titles whose civic reporting is exactly the journalism most worth defending and least worth licensing, because no one is going to pay seven figures a year to cite the Orlando Sentinel in a model response.

That asymmetry is the actual shape of the settlement market. Even if every plaintiff wins, the deals that follow will look like the deals already signed: large conglomerates get cheques, everyone else gets a cease-and-desist letter forwarded to a lawyer who will tell them the cost of litigation exceeds the likely recovery.

The decline was already happening. US newsroom employment fell by about 26 percent between 2008 and 2020, according to Pew Research Center. That number is from before ChatGPT existed. The structural damage to journalism — Craigslist taking classifieds, Google and Facebook taking display advertising, private equity stripping local titles for parts — predates AI by more than a decade. Some journalism economists are right to insist on this, and ORA should not pretend otherwise.

But the fact that AI did not start the decline does not mean AI is neutral in it. AI-generated summaries and answer engines divert the reader traffic that the surviving advertising model depends on. The same models were trained on the journalism whose traffic they now intercept. That is not a coincidence of timing; it is a transfer.

Who actually loses. The journalists whose bylines are in the training data were not asked. The editors who commissioned that work were not asked. The freelancers whose investigations grounded paragraphs of model output were not asked. None of them will see any portion of any licensing payment that flows from these suits, because the contracts under which they wrote assigned the copyright to the publisher. The publisher might cash a cheque; the reporter who lost their job in the last round of cuts will not.

This is the shape of a lot of AI deployment, and it is worth naming clearly. The value of the work flows to whoever holds the contract over the worker who did it. The licensing market that may emerge from these lawsuits is a market between two sets of corporations, large publishers and large AI companies, over a resource (trained-on text) produced by workers who are not party to either side.

The fair use question, taken seriously. OpenAI's defence is that training on publicly available internet text is transformative use, of the kind US courts protected in the Google Books scanning cases. That is not a frivolous argument. The Google Books ruling is real precedent, and the training-versus-output distinction matters: a model that has read a corpus is not the same legal object as a model that reproduces passages from it verbatim.

The trouble is that OpenAI's own models, on the NYT's evidence, did reproduce articles verbatim or near-verbatim when prompted. And the US Copyright Office's 2024 report explicitly said AI training on copyrighted works raises serious questions that existing doctrine does not resolve. The fair use defence may survive in narrowed form. It is unlikely to survive intact.

If it does survive intact, the distributional logic is stark: a century of public-interest journalism becomes a free input to a private system valued in the tens of billions. That is not a fair use story. That is a subsidy story, running from working journalists through the publishers who employed them to OpenAI's shareholders.

What the lawsuits will and will not do. They will probably force a licensing market. They will probably force OpenAI and Microsoft to write larger cheques to a small number of large publishers. They may force some retraining discipline, some attribution requirements, some guardrails against verbatim reproduction.

They will not rebuild local newsrooms. They will not put reporters back into the statehouses and council chambers that AI summaries cannot cover because no one was there to write the original copy. They will not change the contracts under which the journalism that did get written was assigned away from the people who wrote it.

The lawsuits are worth bringing. The fair use question is worth resolving. But the test of whether the journalism industry survives this is not whether courts make AI companies pay. It is whether anyone is still being paid to do the reporting in five years, and the answer to that question does not live in a courtroom.

Footnotes

¶