I have spent the better part of a decade suggesting to news media executives that the regulatory cavalry was not coming. Build the brand, own the relationship, stop waiting for Canberra or Brussels or Washington to fix the platform bargain for you.

I still believe that. But something shifted in the last three weeks, and it’s worth spotlighting.

In the space of roughly 21 days:

• A U.S. federal judge settled the remedies for Google’s ad tech monopoly.

• Meta wrote a cheque for up to US$18 billion with a clause nobody saw coming.

• The Trump administration put its thumb on the scale in The New York Times v. OpenAI.

• Australia closed the loophole Meta drove a truck through in 2024.

• German and European courts told Google it owns what its AI says.

• And the EU switched on the transparency provisions of the AI Act.

That is not six stories. That’s one story: the referees have arrived. The question for every news media company is whether you understand what they are actually refereeing — because it is not journalism, even when the ball lands at our feet.

Google won the battle it needed to win

Start in Alexandria, Virginia. Last week, Judge Leonie Brinkema declined to order Google to sell AdX, the exchange the Justice Department wanted carved out, and opted instead for behavioural conditions on how the ad tech business operates. Google’s regulatory chief called it a win for the tools that help small businesses grow. It was, more plainly, a win for Google.

Let me be precise about what did and did not happen, because it is easy to overstate this one. A break-up was never really in the cards, and it was even less likely after last year’s rulings in the Google search and Meta cases left both companies structurally intact. Brinkema had signalled her practical concerns repeatedly: the years an appeal would consume, the question of who could actually buy and operate AdX at a time when the open Web is shrinking, and the awkward fact that ad tech is a small and rapidly diminishing part of Google’s profit. The behavioural outcome was not a surprise. The anti-trust circus has moved on.

Nor should you assume Google appeals. It got the core outcome it sought — no divestiture of AdX or DFP — and the court adopted most of the remedies the parties themselves proposed, with modifications. One Google proposal we already know about is real-time AdX bid information for rival publisher ad servers. The full package remains sealed for a few more days, and the process of finalising it gets under way this week.

So Google won the battle. But it likely is losing the war, and the war is the one that matters financially to publishers.

Brinkema found illegal monopolies in publisher ad servers and ad exchanges, and she found that publishers were harmed. France’s competition authority earlier found abuse of dominance; Google did not contest the facts, and that decision became final without appeal. The European Commission then fined Google €2.95 billion for abusive ad tech conduct — Google is appealing that one, on grounds unrelated to last week’s finding. Put those together, and Google’s monopoly is now formalised in verdicts across the globe, with the arguments officially established on the record.

That accumulated record matters enormously to the very large publisher damages and class actions now proceeding in many countries and seeking billions. The biggest is in the United Kingdom. Every one of those plaintiffs walks into court with liability already established.

Here is what it means for news media companies in the next 12 months: Nothing structural will change in your programmatic revenue because of a U.S. judge. If you were holding out for a divested AdX to lift take rates, stop — nobody serious was. The behavioural rules will shape how the exchange runs at the margins. The real money is in the damages queue, and it will arrive through litigation, not through Washington. Budget accordingly.

Meta’s clause may be the most important thing that happened in August

The Meta settlement is being reported as a child-safety story. It is also the moment the United States finally imposed practical product constraints on Big Social — and I think that is the bigger story.

Meta settled with 52 attorneys general days into a trial over whether Facebook and Instagram were engineered to hook teenagers. The headline number is up to US$18 billion over 10 years. The surprise is in the fine print. Roughly 70% is unconditional. The remaining 30% — about US$5.3 billion — is payable only if YouTube and TikTok agree to matching payments and adopt the same teen restrictions: a one-hour daily limit, a nighttime blackout, and age assurance.

Meta is now buying full-page ads in newspapers to campaign for its competitors to sign up. Think about what just happened. Meta paid for a legal settlement, used it to take a US$5 billion swing at each of its rivals, and is spending money with publishers to lobby for the standard it wrote.

The mechanics matter. Under-18 accounts on Instagram and Facebook get a default two-hour daily cap, a 12:00 a.m.-to-6:00 a.m. block, muted notifications during school hours, hidden like counts, and an optional non-personalised feed. If YouTube and TikTok join, the cap drops to one hour and the blackout widens. Parents can switch it off. Most will not.

Now put on your audience hat, because this is the main point and not a side effect. If these restrictions become embedded in Facebook, Instagram, YouTube, and TikTok — as Meta is now proposing on behalf of the whole sector — they will almost inevitably be implemented globally or become the template for regulators elsewhere. Product constraints of this kind do not stay inside U.S. borders. That makes this potentially far more important for publishers than the US$18 billion headline.

Consider what follows. If — and it is still an if — the dominant attention platforms in the Western world end up with a one-hour daily cap on teens plus a nighttime blackout, the single largest constraint on youth attention in a generation will have been imposed by state attorneys general, not by product design. Somebody is going to get those hours back. It will not be us by default. It will be us only if we have built something a 16-year-old wants at 10:30 p.m. when Instagram has gone dark.

That is the scarcity thesis in its purest form. Attention is about to become artificially scarce for one demographic. Where does it go?

Washington has picked a side, and it is not the side of the copyright holder

Of the six, this is the genuinely big and unexpected development, and the change of position is the story.

In March, the White House’s national AI framework said two things in one breath:

• The administration believes training on copyrighted material is lawful.

• And Congress should stay out and let the courts decide.

That was the polite version — I have an opinion, but it’s ultimately up to the judges.

Last week, the polite version ended. The Justice Department filed a statement of interest in the Southern District of New York backing OpenAI against The New York Times and the other newspaper plaintiffs, arguing that training large language models on copyrighted text is fair use. It appears to be the first time the federal government has taken a formal position in any of the AI training cases. The brief describes model training as highly transformative, argues the models learn general language ability rather than copying, and frames the whole question as one of scientific progress and national security. The associate attorney general said on X that the administration would not allow the United States to fall behind foreign adversaries because of a mistaken reading of copyright law.

That is a major escalation. We’ve gone from “let the courts decide” to the executive branch actively trying to influence what may become the defining U.S. AI copyright case.

The New York Times’ response was blunt: The government is siding with trillion-dollar AI companies against the creators whose work was taken.

A statement of interest is advisory, not binding. Judge Stein can ignore it. But you don’t need to be a lawyer to read the signal: The executive branch of the largest media market on Earth has framed news copyright as a national-security liability. In the year ahead, that framing will show up in trade negotiations, in pressure on allies, and in every licensing conversation you have with an American AI company. They now negotiate with the government behind them.

My advice has not changed, but the urgency has. If your AI strategy depends on winning in a U.S. courtroom, you don’t have an AI strategy. You have a hope. Licence what you can, protect what you must, and build the provenance layer — cryptographic verification, institutional accountability, brand amplification — that makes your content worth more inside an AI answer than the scraped version.

Australia did the thing Meta said could not be done

Remember March 2024. Meta walked away from its Australian news deals and, later, from news itself. The message was simple: You cannot make us pay for something we do not carry.

Canberra spent two years thinking about that. Two weeks ago, Parliament passed the News Bargaining Incentive. Any platform with a significant search or social service in Australia and more than A$250 million in local advertising revenue now owes a levy of 2.75% of that ad revenue — whether or not it carries a single news story. Deals with publishers offset the levy. The final version narrowed the base from total revenue to digital ad revenue after platform lobbying, added LinkedIn to the list alongside Meta, Google and TikTok, and doubled the weighting for regional, small and under-represented publishers.

So Meta’s choice is now genuinely unhappy. Pay a levy on all Australian digital ad revenue and get nothing for it, or come back to the table with the publishers it publicly divorced and pay roughly the same money for content it claims it does not want. Meta called the draft irrational and discriminatory. The White House called it foreign extortion. Parliament passed it anyway.

One caveat. The rate is settled, but the distribution is not. How the levy money actually reaches publishers is still being finalised and, frankly, it’s a mess. Watch that process as closely as you watched the vote.

For Australian publishers, this is real money in 2027 once the plumbing is fixed. For everyone else, it is a template. Canada, the United Kingdom, South Africa, Brazil, Indonesia — every government that watched Meta walk away from its bargaining code now has a working model for a levy that cannot be dodged by de-platforming news. I would be surprised if at least two copycats were not tabled in the next 12 months.

But be honest with yourself about what this buys. It buys a subsidy. It does not buy distribution, and it does not buy an audience. Meta will pay the levy and continue to starve news of reach, and the publishers who mistake a cheque for a relationship will be the ones still complaining about traffic in 2028.

The courts have decided AI answers are somebody’s words

Of the six, this is the one I would put at the top of a board agenda.

In May, the Regional Court of Munich granted two publishers an injunction against Google over AI Overviews that had wrongly linked them to scams and subscription traps — claims that appeared in none of the sources the AI cited. The court’s reasoning is what matters. A traditional search result is a title, a snippet, and a link to someone else’s words. An AI Overview reads several sources, weighs them, and produces a new text that a normal person experiences as Google speaking. That makes it Google’s own statement, and Google is liable for it. The “users can check the links” defence failed. Google is appealing.

Be clear about what the publishers won, because the distinction matters to anyone who still depends on Google traffic — which is most of us. The injunction bars Google from repeating specific false statements about the two companies. It does not remove them from AI Overviews, and it does not touch their search rankings. That is the good version of the outcome: the publisher stays visible, and the machine stops lying about it. The uncomfortable version is the one to watch for. The court itself noted nothing stops the algorithm generating the same claim again, and the cheapest way for Google to comply with a queue of such orders is simply to leave the complaining publisher out of AI Overviews altogether. A remedy that makes you disappear is not a remedy most publishers can afford.

In June, the Court of Justice of the European Union reinforced the principle that online services are answerable for content they control, rather than merely host. And in July, the UK Jurisdiction Taskforce published its legal statement on AI liability under English law, concluding that defamation liability attaches to whoever counts in law as the publisher of an AI’s output — and that anyone who manually reviews that output before publication is liable as its editor.

Read those together, and a picture emerges. In Europe, an AI summary is a publication, and whoever generates it is the publisher. This cuts two ways for us.

The first way is obvious and good. If Google is legally the author of what AI Overviews say about your company, your journalists, and your subscribers, then you have a tool you did not have before. Two German publishers won an injunction. Others will follow.

The second way is less comfortable. That UK guidance about manual review applies to you, too. If your newsroom deploys AI summaries, AI briefings, AI explainers — and most INMA members now do — you are the publisher in law of what they say. The human-in-the-loop that we all talk about as an ethical nicety is, from now on, a liability shield. Or a liability trap, if the loop is decorative.

Which raises a question I have not heard many CEOs ask: Who, exactly, is the human? Under English law, the person who reviews the AI’s output before publication is its editor and answerable for it. That is a lot to put on an individual. Staff journalists can reasonably expect their employer to stand behind them, as they always have on defamation. Freelancers cannot, and many freelance contracts push the indemnity the other way. If your AI workflow routes machine-drafted text through a contributor for a sign-off, you may have quietly made that contributor the publisher of record. Will journalists want that risk? Some will refuse it, and they will be right to ask what the legal department is offering in return. Sort that out before a court sorts it out for you.

Brussels switched it on

Which brings me to the quietest of the six and, operationally, the most far-reaching. Since August 2, Article 50 of the EU AI Act has been live. The transparency obligations apply immediately to every in-scope system, old or new, and the penalty ceiling is €15 million or 3% of global turnover, whichever hurts more.

The clause that matters for publishers is the one on AI-generated text about matters of public interest — election coverage, legislation, business and health news, the core of what we do. That text must be labelled as AI-generated unless it has been through human review or editorial control, with a clearly identifiable person — natural or legal — holding editorial responsibility. Note the “or legal.” Brussels does not require a byline on the machine’s work; it requires an institution willing to own it. That is lighter than the English defamation position above, and it is precisely the thing a news brand exists to be. The trigger is the date of publication, not generation, so text drafted by a machine in July and published in September is caught. Providers of generative systems already on the market have until December to implement machine-readable marking. Deployers — that is you — had no such grace.

Two things to notice. First, the editorial-control exception is, in effect, a legal definition of what a news brand is. The EU has just told the market that the difference between a labelled machine and an unlabelled publication is a human who takes responsibility. That is the provenance argument written into statute.

Second, don’t assume this is somebody else’s problem because you are not in the EU. The Act reaches everyone whose content is accessed from the European Union. If a reader in Dublin or Düsseldorf can open your AI-drafted explainer, you are in scope — in London, in Sydney, in Houston. The UK has no marking rule of its own, and that changes nothing. Meanwhile, the platforms and a great many others are already building AI-content marks into their systems. The infrastructure for labelling machine text is arriving whether publishers are ready for it or not.

There is a darker reading, and it deserves airing. When GDPR arrived in 2018, more than 1,000 American news sites simply switched Europe off rather than comply, and some stayed dark for months. Expect a version of that here. A regional U.S. or Asian publisher with a marginal European audience and a €15 million ceiling on the downside may conclude the readers are not worth the exposure and geo-fence its AI-assisted content, or its whole site. Is the risk worth it for most foreign news organisations? The honest answer is that the compliance ask is far lighter than GDPR’s — label the text, or put an accountable editor on it — but boards do not always read the statute before they read the fine. If enough of them pull up the drawbridge, the borderless Web we built our digital strategies on becomes a regional one, with a different content regime on each side of every line. That would be a bad outcome for journalism and a worse one for European readers. It is also, at this point, a plausible one.

What the next year actually looks like

Pull back from the six, and look at the shape.

Governments have decided they will regulate the platforms. On children, on national AI advantage, on taxation, news is a bystander to the main event and a beneficiary at the margins. Competition is different, and I want to be precise about it. Publishers were the very customers Brinkema found had been harmed. News Corp, Le Figaro, and Rossel La Voix initiated the French case. The damages litigation now under way across many countries is about recovering economic value extracted from news publishers specifically. On ad tech, we are not bystanders. We are the plaintiffs.

Courts have decided that AI output is speech with an author. That is a genuine shift, and it is the one lever we can pull ourselves without waiting for a minister.

And the American executive branch has decided that news media’s copyright is a strategic inconvenience. That will not change in the next 12 months, whatever happens in the Southern District of New York.

So here is my read on the year ahead:

• Programmatic: No structural change from the remedies, and no appeal to wait for. The money is in the damages queue, led by the United Kingdom.

• Platform payments: Real in Australia once the distribution terms are sorted, plausible in two or three copycat markets, subsidy not strategy everywhere.

• AI training: The American door is closing. The European door is open, but it opens onto a courtroom.

• AI output liability: A new front where publishers have standing, and a new exposure for every newsroom that ships machine-generated text without a named human behind it.

• Youth attention: Possibly, remarkably, about to become scarce by legal settlement — and not only in the United States.

None of this changes the fundamental job. The post-destination era is here, the referees are on the field, and most of the whistles are being blown for reasons that have little to do with journalism — even when the ball lands at our feet. The brands that win will be the ones that used the year to make their content legally, technically, and reputationally distinguishable from the machine.

Which leaves me with the questions I would put to any CEO this month. If Instagram goes dark at midnight for teenagers next year, what will your brand be doing at 12:01 a.m.? If a court asks who is responsible for the last AI summary your newsroom published, is there a name on it — and does that person know it is theirs? And if Washington has decided your archive is a national resource, what have you done to make it a licensed one rather than a scraped one?

The referees have arrived. Whose game are you playing?

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .