The ceasefire that was supposed to end the Persian Gulf war lasted twenty-two days. It did not die in a battle. It died in a wind-down notice, a revoked permit, and a threat level upgraded from elevated to "severe."

Early Thursday, U.S. forces struck Iran for a second consecutive night. U.S. Central Command said the two waves hit more than 170 military sites — air defenses, coastal surveillance, missile and drone infrastructure, naval facilities and Revolutionary Guard assets along Iran's southern coast — with roughly 90 targets in Thursday's round alone. The stated goal was narrow and repeated verbatim in Washington's own statements: to "degrade" Tehran's ability to "threaten freedom of navigation in the Strait of Hormuz."

Iran answered by firing drones and missiles at the three Gulf states that host American power. Sirens sounded at least twice in Bahrain, home to the U.S. Navy's 5th Fleet. Qatar's Interior Ministry declared the "security threat level is high" and told citizens to stay home. Kuwait's military said it intercepted incoming drones and ballistic missiles. As of the latest confirmed reporting, none of the three Gulf states reported damage. Iran's Health Ministry said U.S. strikes across five provinces over two days killed at least 14 people and wounded 78, with 47 still hospitalized.

That is what changed in the last day. What matters more is what didn't: the Strait of Hormuz is not blockaded. It is simply, deliberately, dangerous.

That distinction is the whole story. About a fifth of the world's oil and liquefied natural gas moves through Hormuz. It does not need to be shut to seize the global economy by the throat. It only needs to be frightening enough that shipowners refuse to sail and insurers refuse to write the passage. On that measure, the escalation is already a success — for Iran. Shipping data firm Kpler counted only four tankers clearing the Strait on Wednesday, against a recent daily average around thirty-two. Rystad's read was blunter: navigation had "essentially ground to a complete halt." QatarEnergy turned three empty LNG carriers away from Ras Laffan. An Indian-flagged supertanker carrying two million barrels of Kuwaiti crude U-turned midway through its transit.

The mechanism running underneath all of this is a fight over which lane you're allowed to use. Tehran insists ships transit a northern corridor it controls and can approve. Washington maintains Hormuz is an international waterway Iran has no right to meter. The vessels attacked this week were on the southern, Oman-hugging route the U.S. Navy claims to protect. In other words: Iran is shooting at the ships that refuse to pay its toll, to prove there is a toll.

The economic lever came first, on Tuesday, when the Treasury revoked the general license that had let Iran sell oil on world markets — a central concession of the June 22 interim deal — and gave buyers until July 17 to wind down. Iran's Foreign Ministry called it a "blatant violation" of the memorandum and promised "decisive actions." A day later, President Trump declared the deal itself "over," then told reporters aboard Air Force One that Iran's leadership had called "a little while ago" and "want to make a deal so badly." Tehran has not confirmed any such call, and Trump added he was "not sure" he wanted a deal at all.

So the public sequence is a ceasefire canceled by permit, a permit backed by airstrikes, and a claim of surrender that only one party will say happened. Underneath the noise, the analysts are converging on a colder forecast than the headlines. Clarksons told clients a full closure is not the base case, because "both sides have incentives to avoid a broader oil shock." MST Financial's Saul Kavonic put a number on the alternative: passage "below 50 percent of pre-war levels for many months, with periodic flare-ups."

That is the shape of this war now. Not a front line. A meter. It runs when the toll is paid and stops when it isn't, and everyone — Tehran, Washington, the tanker owners, the traders in Chapter 4 of this paper — has quietly agreed to negotiate over the price of passage rather than the fact of it.

The fragile thing was never the peace. It was the pretense that a waterway this valuable would ever be free.

The cleanest way to misread this week is to say oil spiked because the Strait of Hormuz closed. It didn't close. That is precisely why the move is so instructive.

Brent ripped from below $70 at the start of the month to just under $80 on Wednesday — roughly a 12% re-rate, the fastest since the war began in February. WTI ran the same path from a four-month low near $68.60 to the mid-$70s. On Thursday, with a second night of strikes on the tape, Brent held near $78.80. None of that required a single mine in the shipping lane. It required only that the lane became too dangerous to use casually.

That is the mechanism worth internalizing, because it is the one that will govern this market for months. Hormuz doesn't have to be shut to function as a supply cut. It just has to be frightening. When the naval coalition raised the transit threat to "severe," the deterrence did the work: QatarEnergy pulled three LNG carriers back toward Ras Laffan, a laden VLCC U-turned off Oman, and Kpler counted four transits in a day against a recent norm near thirty-two. A physical chokepoint is a headline. A behavioral one is a tax, and taxes compound.

Strip away the theater and two prices tell you what the professionals actually believe. First, no one is pricing a permanent closure — Clarksons calls a full shutdown a non-base-case because both sides lose in a genuine oil shock. Second, the risk premium is sticky, not spiky, which is the market's way of saying it expects a long stretch of periodic flare-ups rather than one clean catastrophe. MST Financial's Saul Kavonic put the operating assumption in the open: transit "below 50 percent of pre-war levels for many months."

The supply cushion that would normally absorb this is thinner than the calm price action suggests. EIA data show the U.S. Strategic Petroleum Reserve at its lowest since 1983 — the release valve has already been half-turned. And the demand side offers no relief, because the barrels aren't the only thing being metered right now; so is the electricity to cool the country that burns them.

The most expensive misconception in the market this week is that danger and closure are the same trade. They aren't. Closure is a spike you fade. Danger is a floor you pay rent on. This is rent.

The June minutes released Wednesday were the first under new chair Kevin Warsh, and they read exactly like a committee that has lost control of its own inputs. Rates held at 3.5%–3.75% on a unanimous vote. But the projections split the room in half: of eighteen policymakers submitting forecasts, nine penciled a hike before year-end, six of those expecting two quarter-point moves. "Many" officials saw the rate flat or lower by December; "many" saw it higher. That is not consensus with error bars. That is a coin.

What the committee cannot agree on is whether inflation cools once the Iran war winds down or stays hot regardless. And the honest reading of the minutes is that the Fed has outsourced the answer to a waterway it does not control. The June dot plot lifted the year-end PCE inflation projection to 3.6%, up sharply from 2.7% in March — the revision tracks the energy shock, not domestic overheating. Which means the swing variable in U.S. monetary policy this quarter is the toll on a barrel of crude through Hormuz.

Watch the odds move in real time. A December hike was priced near 3% at the June meeting; CME futures now put it around 40%, and September-hike odds climbed to the mid-60s as oil, yields and the dollar rose together this week. The 10-year Treasury sits near 4.58%. Warsh, notably, declined to submit his own forecast — his stated reason is that guidance locks policymakers into a path that's hard to reverse. The generous read is intellectual humility. The market's read is that the chair doesn't want his fingerprints on a number that Tehran gets to revise.

There's a second, quieter worry buried in the minutes that has nothing to do with Iran: officials fret that the AI buildout will keep inflation elevated by bidding up semiconductors and technology goods. Sit with that. The two forces the Fed named as its inflation risks — a Gulf shipping lane and a datacenter arms race — are the exact two stories leading this paper today. The central bank is no longer steering the economy so much as reacting to whichever gate gets throttled first.

Gold, for the record, refused to play its assigned role. It slipped toward $4,050 as the safe-haven bid drowned under higher oil, firmer yields and a stronger dollar. When the hedge against chaos falls on a day of genuine chaos, the tape is telling you the fear isn't war. It's the tightening the war forces.

OpenAI made its GPT-5.6 series — the flagship Sol, the mid-tier Terra, and the low-cost Luna — publicly available Thursday, July 9. The launch itself is routine. How it got here is not.

For weeks, GPT-5.6 shipped only through a staggered preview in which, per Sam Altman's own account to staff, the federal government was "approving access customer by customer." The staging was requested by two offices most people have never heard of: the Office of the National Cyber Director and the Office of Science and Technology Policy. The stated concern is that frontier models have reached an "unprecedented" ability to find exploitable weaknesses in code — the same capability that makes them useful to a security team makes them useful to an attacker. Axios reported the Trump administration cleared the broad launch after technical testing; OpenAI, for its part, cautioned that oversight at this scale "should not become a permanent standard."

The pattern is now the point. Anthropic hit the gate first: Washington ordered it to stop foreign nationals from accessing public versions of its Mythos model, and the company pulled the technology entirely before beginning to restore Fable 5 and Mythos 5 access last week. OpenAI's arrangement was described as more permissive — but it is still an arrangement. A private company's most valuable product now reaches the market through a federal turnstile that opens one customer at a time.

Everything else in tech this week is downstream of that leverage. Both labs are racing to own the silicon underneath the models: OpenAI unveiled Jalapeño, a Broadcom-designed, TSMC-built inference chip already running its Codex-Spark workloads in production tests; Anthropic is in early talks with Samsung on a custom chip using a 2nm process. Both have filed confidential IPO paperwork at valuations approaching a trillion dollars. Terra will launch at half the price of its predecessor as OpenAI moves to lock in customers.

So you have companies chasing a trillion-dollar listing, building their own chips to escape Nvidia, and cutting prices to capture a market — while the actual release date of the product sits in the inbox of a cyber office. The AI industry spent three years arguing about whether the government could regulate frontier models. It turns out the government didn't need a law. It just needed to be the one holding the on-switch, one customer at a time.

The record didn't break, and the reason it didn't is the story.

During the early-July heat dome, PJM Interconnection — the grid serving about 67 million people across 13 mid-Atlantic states and Washington, D.C. — forecast a peak of 166,304 megawatts, which would have topped the all-time summer record of 165,563 MW set in August 2006. On July 2, actual load crested near 162–163 GW. It fell just short. It fell short on purpose.

To get there, the Department of Energy invoked Section 202(c) of the Federal Power Act, issuing emergency orders that let PJM curtail large industrial users — data centers chief among them — and shift them onto backup generation, and that waived certain pollution limits so power plants could run harder. With 160 million people across 30 states under heat alerts, that was the margin between a record and a rolling blackout: a federal switch that dials down the datacenters.

Here is the loop that should worry planners. The AI buildout — the same buildout the Fed named this week as an inflation risk — is a primary driver of the demand growth straining the grid. When extreme heat and datacenter load collide, the fix is to cut the datacenters first. So the electricity feeding the models is now rationed by the same logic rationing the shipping lane feeding the oil: keep the gate open by throttling whoever's willing to pay to keep it open.

PJM ran at more than 90% of capacity, and roughly 9.5 gigawatts of generation was offline in the mid-Atlantic during the peak — including, pointedly, data centers concentrated in Northern Virginia. New York's grid asked customers to conserve; the Midwest's operator flirted with its own record. The heat broke before the grid did. Next time the arithmetic may not be so forgiving, and the country will discover that its two growth engines — cooling and computing — want the same electrons at the same hour.

The number is the provocation. "Michael," Lionsgate's Michael Jackson biopic, has grossed roughly $977 million worldwide, making it the second-highest-grossing film of 2026 behind "The Super Mario Galaxy Movie" ($1.01 billion) and, along the way, the highest-grossing biopic ever made — past "Oppenheimer." With the film still on release in Japan, a billion is a formality.

What the box office bought is a resurrection with a spreadsheet attached. Jackson's catalog streams nearly doubled after the wide release — up 95% from the prior weekend, per Luminate — and his Spotify monthly listeners climbed from around 68 million to 73 million in a week, sitting now near 108 million. For Sony Music, which bought half of Jackson's publishing and masters in a deal valuing his rights at up to $1.5 billion, the movie is less a film than a marketing engine for an asset it already owns. Lionsgate's stock rose. Everyone with a stake in the estate got richer.

And everyone with a memory has to decide what to do with the rest of it. "Michael" is a hagiography of a man whose life ended in litigation and whose reputation never resolved. The film's colossal success is not really a verdict on Jackson; it's a verdict on us — on an audience that will pay a billion dollars for the falsetto and the moonwalk and quietly agree not to bring up the parts that don't sell. This is the biopic as laundering machine: run a contested figure through two hours of soundtrack and choreography, and out the other end comes clean nostalgia you can stream guilt-free.

Compare it to "Michael's" streaming-chart neighbor, Lee Cronin's divisive $90-million "The Mummy" reimagining, which hit No. 1 in 56 countries on HBO Max — a horror film that at least admits it's selling you a monster. The Jackson film sells you a saint and lets 108 million monthly listeners perform the sainthood. The uncomfortable truth Culture keeps circling: audiences don't want their icons examined. They want them exonerated by ticket sales. "Michael" gave them the receipt.

For three years, the debate about governing frontier AI was conducted as if it were a debate about legislation. Should there be a licensing regime? A liability standard? An agency? Congress dithered, the labs lobbied, the think tanks published, and nothing binding emerged.

Meanwhile, the actual regime arrived through the back door, and it works. The government did not pass a law to control GPT-5.6. It simply asked OpenAI to release the model "customer by customer," with two obscure White House offices — the National Cyber Director and Science and Technology Policy — signing off on access one enterprise at a time. Anthropic got the harder version: an order to block foreign nationals from its Mythos model, which it complied with by pulling the product entirely. No statute. No vote. No hearing. Just a request from an agency to a company that decided saying yes was cheaper than saying no.

I want to be fair to the case for this. Frontier models can now find software vulnerabilities at a level that genuinely frightens the people whose job is to worry about it. If a model can hand a hostile actor a working exploit chain, staging its release is not censorship; it's the same logic that keeps certain pathogens out of mail-order catalogs. On the merits of this specific capability, the caution may be correct.

But notice what has been established as precedent, quietly, with everyone's consent. The release of general-purpose software — the defining commercial product of the decade — now runs through a discretionary federal approval that has no enabling legislation, no published criteria, no appeals process, and no sunset. OpenAI itself flagged the danger in the mildest possible language: this "should not become a permanent standard." That is a company that got the green light this time politely reminding the government that next time it might not.

The problem with governing by request is that requests are invisible and unaccountable. A law can be challenged, amended, repealed. A phone call from the cyber director to a CEO cannot. When the gatekeeping is informal, the public never learns which models were slowed, for whom, on what evidence, or whether the same standard was applied to a domestic lab and a foreign one. We are told to trust that the offices exercising this power are wise. Maybe they are. But "trust us, we'll only use it well" is the argument every unaccountable power has ever made, and it has aged badly every time.

Here is the test I'd apply. If the staging of frontier releases is important enough to do, it is important enough to write down — as a rule, with criteria, published, reviewable, and time-limited. If it can't survive being written down, it shouldn't be exercised by a whisper. The most powerful software ever built now ships when a clerk waves it through. We can decide that's acceptable. We just can't pretend we decided it.

We need better words. "Ceasefire" implies a threshold crossed — guns down, a line held, a before and after. What actually happened between the United States and Iran was nothing so solid. It was a subscription, and this week it lapsed.

Look at the structure, not the language. The June 22 general license letting Iran sell oil came with an expiration date — August 21 — like a trial period. When Iran violated the informal terms of service by shooting at ships on the wrong route, Washington didn't declare war. It issued a wind-down notice: transactions must cease by July 17. That is not the vocabulary of statecraft. That is the vocabulary of a lapsed account. The airstrikes were the late fee.

The tell is reversibility. A real peace is expensive to unwind; that's what makes it peace. This arrangement was designed to be canceled. The concession was a license, revocable at will. The enforcement was a strike package that can be launched and stood down in a night. The "deal" was a memorandum of understanding — the diplomatic equivalent of a clickwrap agreement, binding until one party decides the terms no longer suit it. Twenty-two days in, one party did.

This is not an accident of this particular conflict. It is the emerging default. Increasingly, the powerful prefer arrangements they can meter and reverse over commitments they'd have to honor. A revocable permit beats a treaty. A threat level you can raise beats a border you'd have to defend. A subscription beats a sale, because a subscription keeps the leverage with the seller. Iran learned the same lesson from the other side: don't close the Strait, just make it a service you can throttle. Both governments are now running the same business model against each other, and the ships and the barrels and the 78 wounded are the churn.

The cost of the mislabeling is borne by everyone downstream who plans as if peace means stability. The trader who priced Hormuz back to pre-war levels. The shipowner who sent a laden VLCC into the corridor. The reader who saw "ceasefire" in June and exhaled. They were all reading a subscription as a settlement.

So let's stop. When a government offers a rival a deal that expires on a date, backed by a concession it can revoke on a whim, that is not peace being made. That is a meter being installed. Call it what it is, and at least you'll know to expect the bill.