Washington spent Wednesday preparing for a signature and Thursday absorbing a delay. The planned White House ceremony for President Donald Trump's artificial intelligence and cybersecurity order was postponed, Axios reported, after Reuters said the order could come as soon as Thursday. That may sound like scheduling weather. It is not. The ceremony slipped; the governing premise did not.

The draft framework, as described by Reuters and Axios, would ask leading AI developers to engage with the government before releasing covered frontier models. Axios reported that the current draft contemplates labs sharing models with the government at least 90 days before public release and giving access to some critical infrastructure providers. The word voluntary is doing a great deal of labor here. In practice, a voluntary lane endorsed by the White House can quickly become the only lane that serious companies, cloud providers, banks, and defense contractors are allowed to travel.

The trigger is not an abstract fear of smart chatbots. It is cyber capability. Anthropic's Claude Mythos has become the model Washington cannot stop talking about: restricted, dangerous enough to unsettle security agencies, useful enough that nobody wants to be left outside the room. The company is briefing the Financial Stability Board, according to the Guardian, after giving selected banks and technology companies access for defensive testing. Scientific American reported that Mythos is being kept out of public release while a small group uses it to find flaws before attackers can. That is not normal product launch behavior. It is more like controlled handling of a tool that can see too far into the machinery.

This is why the story belongs on the front page rather than the tech page. The state is trying to invent a chokepoint for cognition. It wants sight of the model before the market gets it, partly to protect hospitals, banks, and software infrastructure, and partly because no administration wants to explain after the fact why it learned about a national-scale vulnerability from an earnings call.

The foreign policy half is already attached. Trump said after meeting Xi Jinping that he discussed AI guardrails and Nvidia's H200 chips with China. Nvidia has U.S. licenses for H200 sales, but Reuters reported there was no immediate Chinese approval. Taiwan prosecutors, meanwhile, are investigating alleged smuggling of high-performance AI servers containing Nvidia chips into China. The rulebook and the black market are growing at the same time.

The old AI debate asked whether machines could think. The new one asks who gets advance notice when they can break things. That is a colder question, and a more serious one. The demo era is being replaced by a permissions era. The public may still meet AI as an app. Governments are beginning to meet it as infrastructure, leverage, and liability.

The European Commission's new forecast is not a panic note. It is worse: a spreadsheet that has calmly accepted the war. The commission now expects the eurozone to grow 0.9 percent this year, down from a prior 1.2 percent forecast, and sees inflation at 3.0 percent, up from 1.9 percent. The difference is not rhetorical. It is the cost of energy, shipping anxiety, and consumers who can read a gas bill as quickly as a headline.

AP reports that the commission still expects the euro area to avoid recession. That is the soft landing version of bad news: people keep moving, firms keep hiring selectively, and central bankers keep looking at prices that refuse to behave. The Strait of Hormuz is the hinge. In normal times, roughly a fifth of the world's oil and natural gas travels through it. In this one, risk of Iranian drone and speedboat attacks has closed off most ship traffic.

The economic fact is simple and unsentimental. Europe tried to turn the page from one energy emergency, only to discover that the next one could be printed on shipping maps rather than pipeline maps. Higher oil prices do not need to become 1970s mythology to bite. They need only last long enough to affect transport, chemicals, fertilizer, and expectations.

The commission's downside warning is where the politics lives. A prolonged period of higher energy prices would push growth lower and inflation higher. That means the Iran war is no longer a foreign-policy column. It is a wage negotiation, a rent conversation, a mortgage-rate argument, and a test of how much pain voters will absorb before they demand someone, somewhere, make the sea lane ordinary again.

The House vote expected Thursday is not merely about Iran. It is about whether Congress can still hear itself over the machinery of a war already underway. AP reports that lawmakers are set to consider legislation compelling President Trump to withdraw from the conflict, a war the president launched more than two months ago without congressional approval.

That phrase, without congressional approval, should still clang. It has become too familiar in American life: presidents move first, legislators posture second, courts arrive late if they arrive at all, and the public is left to sort out the difference between support for troops and consent to policy. The Iran war has made the old constitutional argument concrete again because its costs are not theoretical. They are visible in oil markets, inflation forecasts, and the diplomatic scramble around the Strait of Hormuz.

Democrats are hoping for a breakthrough. The more interesting question is how many Republicans are willing to make one possible. War powers fights often become partisan theater until they threaten a president from the lawmaker's own party. Then the institutional principle is asked to survive contact with the whip count.

The vote may fail. Many important votes fail before they matter. But every failed vote builds a record of who treated war as an executive convenience and who treated it as the gravest power the republic can grant. If Congress wants its constitutional muscles back, it has to use them while the conflict is live, not after the memoirs arrive.

The midterms are not waiting for autumn. They are already moving through databases. AP reports that at least 67 million voter registrations, largely from Republican-controlled states, have been checked through an expanded Homeland Security verification program, with tens of thousands flagged as potential noncitizens or deceased voters.

The numbers require proportion. Citizenship and Immigration Services said 60 million checks identified about 24,000 potential noncitizens. Officials also cited hundreds of thousands of records that may belong to deceased voters. Even if every flag were right, which voting-rights lawyers strongly dispute, the shares would be small. That does not make the system harmless. Small error rates become real people when states give voters little time to prove eligibility or suspend registration immediately.

The fight is not just technical. It is ideological. The administration frames the program as election hygiene. Critics call it a midterm purge machine, one built on incomplete records and deployed in an atmosphere where the president has repeatedly promoted claims about noncitizen voting despite rare documented instances.

The legal ground underneath is also shifting. AP separately reported that the Supreme Court has sharply weakened Section 2 of the Voting Rights Act, narrowing one of the main tools used to challenge minority vote dilution. Put those developments together and a pattern emerges: less judicial protection against discriminatory maps, more executive pressure on voter rolls, and a midterm map increasingly shaped before voters see a ballot.

American democracy now has a user-interface problem. The line at the polling place is visible. The database query is not. A voter wrongly flagged may not know the injury until the election has become an artifact.

Nvidia delivered what used to pass for magic: stronger profit, stronger revenue, and a forecast above analysts' expectations. The stock still slipped. AP reported that shares moved between gains and losses before falling 1.9 percent, while the broader market drifted under pressure from oil prices and Treasury yields.

That is not a rejection of the AI boom. It is the market developing manners. Reuters noted that Nvidia showed almost a doubling of sales over the past year, lifted its dividend, and planned an $80 billion buyback. There was no obvious crack in demand. Chip stocks abroad rallied. The problem is that investors have already built a cathedral around the company and now inspect each brick for divine properties.

The larger market tells the same story in duller numbers. Brent crude climbed above $107 a barrel in AP's Thursday report as uncertainty around the Strait of Hormuz persisted. The 10-year Treasury yield rose to 4.60 percent. Higher yields do not merely compete with equities; they raise the cost of building the data centers that have become the capital-intensive altar of the AI trade.

The bull case remains powerful. The infrastructure buildout is real, corporate spending is not imaginary, and software will continue to eat the office with better table manners. But the easy phase is over. When a company can beat expectations and receive a shrug, the story has moved from discovery to duration. The market believes in AI. It is now asking how much belief can be financed at 4.6 percent.

The semiconductor war is usually described in grand strategy language: sovereignty, deterrence, technological leadership. The latest case is smaller and more revealing. Taiwan prosecutors are investigating three people accused of conspiring to buy high-performance Super Micro AI servers in Taiwan, use false export documents, and smuggle them to China, AP reported.

The case sits neatly inside the contradiction of the year. Washington wants to control China's access to advanced compute. Beijing wants domestic champions strong enough to avoid dependence on American chips. Companies want revenue, customers want capacity, and somewhere in the middle a customs form becomes a geopolitical instrument.

Nvidia's official China channel is already jammed. Reuters reported this week that Jensen Huang believes the Chinese market will open over time, but Nvidia has U.S. licenses for H200 chip sales and still lacks Chinese approval. Trump said after his Beijing summit that H200s came up with Xi Jinping. China, according to the president, wants to develop its own.

That is the policy layer. The smuggling allegation is the market layer. When demand is large, restrictions are uneven, and chips are both commodity and strategic asset, enforcement moves from export-control memos to warehouses, forwarders, and prosecutors. The future of AI may be announced from keynote stages. Its limits will be tested in ports.

Germany is floating a phrase designed to sound smaller than history: associate membership. Chancellor Friedrich Merz wants the European Union to consider that status for Ukraine and to revive talks aimed at ending more than four years of war with Russia, according to a letter seen by AP.

The offer is not full EU membership. That is precisely why it matters. Full accession is slow, bureaucratic, and politically combustible. Associate membership would give Kyiv something less than the dream but more than the waiting room. In any peace process where Ukraine may be pressured to accept losses it cannot celebrate, a formal European track could become part compensation, part guarantee, part domestic argument.

This is the diplomacy of survivable disappointment. Ukraine wants security and belonging. Russia wants constraints on Ukraine's future. Europe wants the war ended without rewarding conquest so openly that every border on the continent starts to look provisional. No formula solves that cleanly.

Merz's proposal recognizes that peace talks are not only about territory and guns. They are also about what a country can tell itself afterward. If NATO remains blocked and full EU accession remains distant, an associate door may be the thing Europe can open now. It is not enough. It may be more than symbolism.

The most useful thing Cannes can do this year is disappoint people looking for Hollywood's reflection. The Guardian reports that the 2026 lineup has a near-total absence of major Hollywood studio films, with the competition leaning heavily into international auteurs: Asghar Farhadi, Laszlo Nemes, Cristian Mungiu, Andrey Zvyagintsev, Hirokazu Kore-eda, Ryusuke Hamaguchi, and others.

This is not austerity. It is taste reasserting itself after years in which every festival became a launchpad, content bazaar, award-season weather system, and luxury step-and-repeat. Cannes is still glamorous, absurd, and perfectly capable of mistaking a staircase for civilization. But the thinner studio presence gives the festival back a more interesting job: making difficult cinema feel like the center of the world for eleven days.

The old Hollywood aura has not vanished. Barbra Streisand is still due to receive an honorary Palme d'Or, though AP reports she will not attend the closing ceremony because of a knee injury. Peter Jackson received an honorary Palme at the opening, and John Travolta was surprised with one ahead of his directorial debut. The statues keep arriving; the studios less so.

That imbalance is the point. A festival does not become important by hosting the biggest machines. It becomes important by making smaller, stranger, more stubborn work impossible to ignore. If Hollywood wants the Croisette back, it can bring films that need judgment rather than merely attention.

A reader wrote us this week with a sharper point than most market notes: oil can be cheap because growth is no longer simply an oil economy. The instinct is right. The marginal story of rich-world growth increasingly runs through data centers, chips, software labor, electricity, and capital expenditure that looks more like grid planning than refinery output.

But the Iran war is teaching the cruel addendum. Growth may be leaving oil faster than oil is leaving growth. The European Commission just cut eurozone growth expectations and raised inflation forecasts because energy prices rose with the Strait of Hormuz crisis. U.S. stocks are moving around Brent crude and Treasury yields while Nvidia, the icon of the new economy, finds that even excellent earnings cannot fully overpower old-economy pressure.

The mistake is to treat oil relevance as binary. Either barrels rule everything or they are yesterday's concern. Reality is uglier. AI shifts the composition of demand; it does not abolish logistics, plastics, aviation, shipping, military fuel, fertilizer, or the psychological power of a gasoline sign. Data centers run on electricity, but electricity is still built from hardware, land, cooling, turbines, transmission, debt, and permitting. Much of that world remains exposed to energy prices, interest rates, and war.

So yes, the reader was right to push us harder on AI. The economy's growth engine is changing. That is exactly why the oil shock matters in a new way. It is no longer the whole machine seizing up. It is the old bottleneck taxing the new machine while pretending to be obsolete.

There is a case for the government seeing the most dangerous AI models before everyone else does. It is not a weak case. If a model can accelerate cyber exploitation against banks, hospitals, browsers, and operating systems, public release cannot be treated like a phone update with better adjectives.

But the emerging White House framework has a democratic problem hiding inside its practical appeal. Voluntary pre-release sharing, CEO ceremonies, selected critical infrastructure access, and agency review may create a faster safety channel. They also risk creating a velvet rope system in which the future is negotiated among labs, regulators, banks, cloud giants, and defense-adjacent institutions before the public knows what bargain has been struck.

That is not a constitution. It is a guest list.

AI oversight needs speed, but speed cannot be the only virtue. The public deserves clear thresholds for what counts as a covered frontier model, firm rules for what the government may do with pre-release access, and disclosure after emergency secrecy expires. Companies need a predictable process rather than a favor economy. Smaller firms need to know whether voluntary means optional or merely optional until procurement officers punish nonparticipants.

The postponement of the signing ceremony is an annoyance, not a tragedy. The delay should be used to make the framework less theatrical and more legible. The state can have an early-warning system without turning AI governance into court politics. If the models are now infrastructure, the oversight must be public architecture, not a private hallway.