The United States government entered a copyright case this week to argue that an artificial-intelligence company should generally be allowed to ingest published journalism without paying for it. Hours later, a federal judge declined the same government’s request to break apart Google’s advertising machinery after finding that the company had illegally monopolized important parts of the market through which publishers earn money.

These are separate proceedings, involving different legal questions and different branches of government. They nevertheless describe one economic system with unusual clarity. At the front door, the state says the machine’s consumption of a newspaper may constitute fair use because artificial-intelligence development advances national prosperity and security. At the back door, a court says the dominant intermediary may keep the exchange through which that newspaper tries to finance itself, subject to behavioral changes that remain sealed for two weeks.

The Justice Department’s statement of interest supports OpenAI in the copyright lawsuit brought by The New York Times and other publishers. It does not decide the case. The department told the Manhattan federal court that treating the use of copyrighted work for model training as infringement could obstruct scientific progress, economic mobility and American leadership. The Times answered with the old commercial proposition in freshly urgent form: creators and AI companies can coexist if the companies pay for the material their products require.

In Virginia, U.S. District Judge Leonie Brinkema spared Google from a forced sale of its AdX exchange. She had previously found that Google illegally maintained monopolies in two publisher-side advertising-technology markets. Her short order accepts modified behavioral remedies while withholding the detailed opinion temporarily for redaction. The government had argued that leaving the stack intact would permit Google to manipulate algorithms in ways too difficult to detect. Google welcomed the rejection of a breakup.

A newspaper therefore confronts two versions of scale. One scale makes copying so comprehensive that each copied work is presented as merely an ingredient in transformation. The other makes market control so embedded that removing the machinery is judged more disruptive than regulating its owner’s conduct. The publisher remains recognizable enough to sue but apparently not important enough to disturb either system.

This is not an argument that every model-training use must be unlawful, nor that every antitrust violation requires corporate dismemberment. It is an accounting problem. Journalism costs money before it produces a word worth ingesting: reporters call people who do not answer, lawyers review allegations, photographers travel, editors discard weak drafts, and corrections preserve a record after publication. Neither model weights nor advertising exchanges naturally reimburse that work. They optimize after it exists.

The national-interest argument also carries an asymmetry. If the production of reliable information matters to public safety, scientific progress and democracy, then the institutions that produce it belong inside the national-interest calculation. Washington’s filing describes the strategic value of models trained on a rich information commons. It says much less about who will pay to keep that commons rich.

The most consequential technology policy is often not called technology policy. It is a copyright brief defining permissible intake, an antitrust remedy defining tolerable control, or a procurement rule deciding which company becomes infrastructure. Wednesday supplied two of those decisions at once.

For publishers, the warning is not that machines are coming. They arrived years ago. The warning is that the legal settlement now forming may treat original reporting as a free strategic input while treating the systems monetizing attention as too integrated to separate. A country can choose that arrangement. It should at least be required to say the whole sentence aloud.

More than 800 students and teachers in Indonesia became ill this week after eating meals supplied through the government’s free-food program, according to reports from three separate incidents in Java and Sumatra. Images from Sidoarjo showed children receiving treatment after food distributed at school made them sick.

The program is a flagship undertaking built around an uncomplicated promise: feed children, improve nutrition and relieve pressure on household budgets. The illnesses expose the complicated machinery beneath it. Every additional kitchen, delivery route and supplier expands the number of places where temperature control, storage, preparation or supervision can fail. A benefit delivered at national scale is also a national food-service operation, whether politicians describe it that way or not.

The relevant measure is no longer only how many portions the government can distribute. It is how quickly authorities identify a contaminated batch, how precisely they trace its ingredients, whether separate incidents share a supplier, and how transparently the findings are published. Enrollment figures make a program sound large. Traceability makes it safe.

School-meal programs can be unusually valuable because they reach children through infrastructure families already use. That same concentration raises the cost of error: one faulty preparation process can expose an entire school before the first patient reaches a clinic. When teachers are sickened too, the institution responsible for recognizing and managing the event loses staff at the moment it needs them.

Indonesia’s government now has a chance to demonstrate that public provision includes public inspection. It should publish incident-level counts, kitchen and supplier identities, laboratory results, corrective actions and reopening criteria. Free food is not free if families must price in the risk of the hospital.

A federal judge in Maryland blocked President Donald Trump’s latest attempt to restrict birthright citizenship Wednesday, issuing a preliminary injunction while a class-action case brought by immigrant families and advocacy organizations proceeds.

The revised executive order was presented as a narrower measure aimed partly at “birth tourism.” It sought to deny automatic citizenship to several categories of children, including some born to adults deemed alien enemies or connected to what the order described as a commercial transaction to obtain citizenship. U.S. District Judge Deborah Boardman found that the narrower vocabulary did not cure the underlying problem. Children in the certified class, she wrote, are citizens at birth under the Supreme Court’s ruling.

Birth tourism is already grounds for visa scrutiny when officials determine that a person seeks entry principally to give birth. The new order attempted something different: changing the status of the child after birth according to the government’s characterization of a parent’s travel, affiliations or intent.

That distinction matters. Immigration enforcement governs admission and removal. Citizenship determines membership in the polity. Allowing an executive agency to convert disputed facts about a parent into the loss of a child’s constitutional status would place citizenship documents downstream of administrative suspicion.

The administration’s original, broader order was defeated at the Supreme Court in June. Its August replacement tested whether a smaller set of labels could reach a result the constitutional rule forbids at larger scale. Boardman’s answer was no.

The injunction is preliminary, and the class-action litigation continues. But the morning’s operative fact is straightforward: agencies may not enforce the new restrictions against the covered children. The government changed the categories. The constitutional sentence survived them.

Brent crude eased 0.4 percent to about $95.26 a barrel in Asian trading Thursday, while U.S. crude held near $90.84 and major Asian equity indexes advanced. This was described, correctly, as a calmer session. The adjective now deserves inspection.

The retreat followed gains driven by renewed fighting between the United States and Iran. President Trump said Wednesday that he did not expect the bombing campaign to last much longer, and markets accepted a small reduction in immediate risk. They did not restore the missing physical margin.

U.S. Energy Information Administration figures cited by Rabobank showed commercial crude stocks falling 4.5 million barrels last week and the Strategic Petroleum Reserve losing a little more than 3 million. American refineries were operating at 98 percent of capacity. Gasoline inventories fell 1.2 million barrels, while shipping disruption around the Strait of Hormuz and attacks on Russian refining infrastructure kept product markets tight.

A refinery system at 98 percent cannot answer a new shortage with another large turn of the dial. It can defer maintenance, bid harder for crude, alter yields at the margin and hope every essential unit remains online. None of those is spare capacity.

That is the distinction between a price pause and a supply resolution. Traders can mark down the probability of another strike before breakfast. They cannot mark up refinery throughput beyond the physical plant, reopen a contested shipping lane by spreadsheet or rebuild strategic stocks with a futures contract.

Equities received the headline they needed: oil stopped rising for a session. Consumers received a different reality: gasoline remained above $4 a gallon through August, refining margins widened, and the system entered September with little mechanical room for error. Ninety-five-dollar Brent is not cheap. It is simply yesterday’s emergency with a minus sign in front of it.

OpenAI says its forthcoming Astra model is the first of its systems to reach the company’s “Critical” cybersecurity capability threshold: the level at which a model can independently discover previously unknown software vulnerabilities and turn them into working attacks against hardened systems.

During testing, the company says Astra discovered and chained two zero-day vulnerabilities. OpenAI still plans a broader release, but access to its most powerful cyber capabilities will initially be limited to trusted testers and selected organizations. The company also warns that its safeguards may interrupt legitimate security work when they mistake authorized activity for abuse.

The announcement matters less as a benchmark trophy than as a governance event. Capability frameworks are easy to publish while every model remains below their brightest red line. Their credibility begins when a commercially valuable system crosses it.

OpenAI’s chosen response is containment rather than cancellation: restrict users, monitor activity and separate ordinary access from advanced cyber functions. That may prove defensible. It also transfers much of the safety case into operational questions the public cannot answer from the threshold label alone. Who qualifies as trusted? Which capabilities are technically separable? What logs are retained? What event triggers suspension? Who is told when safeguards fail?

There is a second constituency: defensive researchers. A system able to find unknown flaws can help repair widely used software before attackers exploit it. Excessive filtering could obstruct that work; permissive access could industrialize it for adversaries. The difficulty is real, but difficulty is not a substitute for published criteria.

Astra’s release will establish the practical meaning of “Critical.” If the designation produces only a smaller invitation list and stronger warnings, it is marketing taxonomy. If it produces auditable access controls, independent testing and explicit stopping conditions, it becomes governance. The model has crossed the threshold. Now the framework must show whether it can.

The 83rd Venice International Film Festival opened Wednesday with Danny Boyle’s “Ink,” a film about Rupert Murdoch, and a lifetime-achievement Golden Lion for George Clooney. Among the 21 films competing for the festival’s top prize is “NAZA,” a documentary focused on Gaza. Maggie Gyllenhaal chairs the jury, which will announce awards on September 12.

Venice has therefore begun not merely with stars but with rival theories of the image. One concerns the proprietor who accumulates the power to decide which facts become spectacle. Another concerns the camera pointed toward a war whose images are endlessly distributed and endlessly contested. Looming above both is the red carpet, the festival’s own machine for converting attention into rank.

Journalism is irresistible material for prestige cinema because it supplies ready-made moral tension: access versus independence, publication versus silence, truth versus the institution paying the expense account. It also lets films flatter their audiences. Viewers can identify with the brave reporter while remaining comfortably outside the newsroom’s payroll, deadlines and compromises.

The sharper question for this festival is whether these films treat journalism as heroic décor or as an industry of choices. Murdoch’s power was not only personality; it was ownership, distribution and the ability to make political actors anticipate a headline before it existed. A Gaza documentary cannot be assessed solely by the urgency of its subject; it must also answer whose testimony structures the film and what the edit excludes.

Venice is an ideal place for that argument because festivals are editors with tuxedos. They select, sequence, headline and confer authority. The Golden Lion is not truth, any more than a front page is. Both are declarations that amid an excess of available images, somebody chose this one and accepts responsibility for the frame.

Google was found to have illegally monopolized two publisher-side advertising-technology markets. The government sought divestiture. The court declined and ordered behavioral changes whose details will remain sealed for two weeks.

This is the American antitrust ritual in miniature: establish that a corporation acquired or protected power unlawfully, recoil from altering the structure that embodies the power, then issue rules requiring that corporation to behave differently while preserving its incentives, personnel, data and integration.

Behavioral remedies are attractive because they appear precise and moderate. They can prohibit a contract term, require access or constrain a particular transaction. They also demand a regulator who understands a moving technical system as well as the engineers paid to redesign it. The government warned that Google could manipulate the algorithms driving its monopolies in ways too difficult to detect. That was not an argument for a cleverer chore chart. It was an argument that supervision is structurally mismatched to the object being supervised.

Divestiture carries costs. So does allowing an adjudicated monopolist to retain the machinery through which the violation occurred. Courts tend to describe the first category vividly—disruption, complexity, damage to customers—and the second abstractly, as an implementation risk. Publishers experience that abstraction as revenue.

The sealed opinion may contain formidable restrictions. We will read it when it appears. But the governing presumption is already visible: corporate integration is treated as a fact of nature, while competitive markets are an aspiration to be reconstructed through compliance.

A remedy should change the expected return from breaking the law. If the company keeps the strategic asset, absorbs monitoring and continues operating at immense scale, the lesson to every adjacent platform is not restraint. It is endurance. Become sufficiently indispensable and punishment will arrive as product requirements.

The Justice Department says a ruling against OpenAI’s use of published material for model training could hinder American prosperity, science and national security. Accept the premise for a moment. High-quality text is a strategic resource. Its availability improves machines the government regards as essential to national power.

Then follow the premise one step backward. Strategic resources require replenishment.

A newspaper archive does not descend like rain. Somebody attended the hearing, obtained the document, called the witness, checked the number and accepted legal responsibility for the sentence. Model developers call the resulting corpus “publicly available” because that describes where they found it. It does not describe how it came to exist.

The fair-use question belongs to the court, and transformation is a genuine principle rather than a trick invented by AI companies. But the national-interest argument is political. It asks a judge to weigh the future of an industry the government favors while leaving the future of the input industry to subscription cancellations, philanthropic fashion and an advertising market dominated by platforms.

If advanced models require continuing access to trustworthy reporting, policymakers have choices: collective licensing, bargaining rights, research levies, tax credits for original reporting, public-interest procurement or a statutory compensation system. Each has defects. All are more honest than declaring the output strategic and the input incidental.

Publishers also have obligations. Payment cannot become a license for weak reporting or a cartelized right to freeze new tools. News organizations should disclose AI agreements, preserve human accountability and show readers what their reporting costs. The answer to extraction is not mysticism about ink.

But a nation cannot build information machinery by liquidating the institutions that gather information and call the result progress. Every strategic model needs current, contested, expensive facts. Somewhere behind those facts is a person with a byline. Put that person in the national-interest brief.