The Tenth Vote

Two crises. Two different worlds. The same broken machinery underneath.

"Those who are to conduct a war cannot in the nature of things, be proper or safe judges, whether a war ought to be commenced."

— James Madison, 1793

On Tuesday, the United States Senate voted for the tenth time this year to try to stop a war.

It finally worked. Fifty votes to forty-eight. Four Republicans crossed party lines. The resolution directed President Trump to remove American forces from hostilities with Iran, the first time both chambers of Congress had passed a resolution ordering a president out of a war under the War Powers Act.

Then nothing happened.

It's a concurrent resolution. It doesn't go to the president's desk, doesn't require his signature, carries no force of law. A White House official dismissed it within hours, calling it of no significance. The war continues. The Senate's tenth attempt to assert its constitutional authority over war succeeded completely, and changed nothing at all.

Congress voted ten times to stop a war it never declared. On the tenth try, it won. The president shrugged.

This has nothing to do with whether you supported the war. It has everything to do with how power actually moves now. The pattern underneath the Iran vote is the same pattern we covered in Issue Two, when a single Commerce Department letter shut down two of the most powerful AI systems ever built, at 5:21pm, on a Friday, with no hearing, no vote, no appeal.

Different domain. Different president. Same disease.

What Actually Happened in Iran

On February 28, 2026, the United States joined Israel in strikes against Iran, killing senior officials including the country's leader, Ali Khamenei. Iran retaliated within days, firing missiles at Israel, American bases, and US-aligned Arab states, and closed the Strait of Hormuz, through which roughly a fifth of the world's oil and gas passes. The fuel crisis reached gas pumps in Spokane within weeks.

The war ran nearly four months. CSIS puts the Pentagon's cost at roughly $40 billion. Gas prices rose from under $3 a gallon to well over $4.

On June 17, the US and Iran signed a memorandum of understanding at Versailles, setting a 60-day clock for a final settlement. Fighting in Lebanon continues. Whether the deal holds is still unclear.

Here is the part that matters most. Congress never declared this war. The Constitution gives that power to Congress alone. The president acted first. Congress spent four months catching up to a decision it never made.

Ten Votes, and What They Actually Proved

The number ten is worth sitting with.

Most attempts failed outright. The House passed its version June 3, 215 to 208. The Senate finally passed its own June 23, fifty to forty-eight.

Senator Tim Kaine put it plainly. The most solemn power Congress holds, he said, is the power to declare war. Not the president. He's right about what the framers understood. He's also describing a power that no longer functions that way.

The War Powers Resolution already requires the president to notify Congress within 48 hours of sending troops into hostilities, and requires withdrawal within 60 days absent approval. That's been law since 1973.

None of it stopped this war, or its continuation past that deadline, through a maneuver in which the administration simply declared hostilities terminated by a ceasefire that, on the ground, kept being violated.

The law exists. It has for fifty years. It no longer works, because executive action now moves faster than Congress can object.

Yale's Oona Hathaway notes the last president to actually ask Congress for a formal declaration of war was Franklin Roosevelt, after Pearl Harbor. Every conflict since has run on something looser, an authorization, a claim to inherent defensive power stretched a little further each time.

Matthew Waxman calls the design "ambition checking ambition," Madison's phrase for how the system was meant to work. His read: presidents keep asserting that ambition. Congress mostly doesn't push back.

Ten votes isn't evidence the system works. It's evidence of how hard Congress had to fight, against its own party discipline, for one symbolic gesture with no legal force.

The Same Shape, Wearing Different Clothes

Now place this next to the AI story from two issues ago.

On June 12, the Commerce Department sent Anthropic a letter at 5:21pm, on a Friday, ordering it to cut off Fable 5 and Mythos 5 for every user it couldn't verify as American. Since Anthropic had no practical way to check citizenship at that scale, the models went dark for everyone.

The stated justification was a jailbreak, one narrow workaround in one case. Anthropic found the same capability already sitting unrestricted on a competitor's model. No hearing. No appeal. No consistent standard.

What both stories share, underneath the surface: one part of the executive branch acted unilaterally, without waiting for the body assigned to check it. That body, Congress in one case, regulatory due process in the other, was structurally too slow to act before the decision was already in effect. When the check finally arrived, it changed nothing. The war continued. The export control stood. And the public justification didn't survive close inspection: the AI threshold was already crossed by an unrestricted competitor, and the war's rationale shifted weekly, from preventing a nuclear weapon, to destroying missile capability, to regime change.

This isn't a partisan point about one president. It's a structural point about what happens when the machinery built to check power moves at a fundamentally different speed than the power it's meant to check.

Why Self-Regulation Keeps Almost Working

There's a comforting story told about both domains: powerful institutions can mostly govern themselves, and formal checks only need to step in as a backstop.

It's comforting because it's sometimes true. Presidents have gone to Congress before war, George W. Bush did, in 2001 and 2003. AI companies have built real safety infrastructure before being forced to. Anthropic spent thousands of hours red-teaming Fable 5 with US and UK safety institutes before its June 9 launch, three days before the shutdown letter.

Watch what happens the moment real pressure hits.

Anthropic built genuine safety infrastructure in good faith. It got shut down anyway, on a rationale its own review showed was applied selectively. This administration, like others before it, has leaned on creative legal arguments, a ceasefire here, an inherent authority there, to avoid asking Congress the question the Constitution requires.

Self-regulation works fine in calm weather. It fails exactly when the stakes are highest, which is exactly when you need it most. That's not coincidence. That's the predictable behavior of any system where the actor and the judge of the action are the same hand.

Madison saw this coming two centuries before Anthropic or the Commerce Department existed.

Those who conduct a war, he wrote, cannot be safe judges of whether it ought to be commenced. Swap in any high-stakes, fast-moving decision. The logic holds. The official who shuts down an AI model is not a safe judge of whether that shutdown is warranted or consistent. He is, structurally, in exactly Madison's position. The actor judging his own action.

What Actually Fixes This

In Issue Two, we laid out five principles for AI governance. Two apply just as cleanly to war powers as to export controls.

Transparency Before Action. No government should restrict an AI system, or commit forces to hostilities, without a public, timely accounting of the finding and authority behind it. Commerce gave none. Congress eventually got an accounting on Iran, months after the fact, when it could no longer change anything.

An Independent Technical Tribunal. Whether a jailbreak justifies shutting down a model used by hundreds of millions shouldn't be one bureaucrat's call on a Friday afternoon. Whether a foreign threat justifies combat shouldn't be one president's call either, defended after the fact with whichever legal theory holds up that week. Both decisions need a body empowered to evaluate the claim before the action, not after the war is four months old.

Ambition must be made to counteract ambition. Madison wrote that in 1788. We're still waiting for the ambition that counteracts ours.

Here's the harder point. Congress had the legal tools to contest the Iran war for months. It took ten votes and four months to use them, and even then the result carries no legal weight. If our oldest, most carefully designed check on power struggles this badly against a war, a domain we've regulated for two centuries, what does that say about our odds with AI? AI moves faster than troop deployments, scales faster than missile production, and diffuses across borders faster than any military technology in history.

If the checks on war are this slow, the checks on AI cannot be modeled on them. We need governance built for the speed of the thing being governed, not the speed of the institutions doing the governing.

The Real Stakes

Forty billion dollars. Nearly four months. A fuel crisis that reached gas pumps in every American city. That's the price of a check on power that arrived too slowly to matter.

AI governance carries the same disease, in a technology that moves faster and touches more lives per day than any war ever has. The Friday letter was a small preview. The tenth vote was a larger one. Neither was the last.

Madison built a system on the belief that no single actor should be trusted to judge the wisdom of his own action. He was right about the principle. He couldn't have anticipated a world where the actor moves at the speed of a press release and the check moves at the speed of a fifty-vote majority, four months late.

The work ahead isn't choosing a side on Iran, or deciding whether Anthropic or Commerce was right. The work is building checks that move at the speed of the century we live in. Until we do, every Friday at 5:21pm, and every tenth vote, arrives after the decision that mattered has already been made.

JOIN THE CONVERSATION: Where else do you see power moving faster than the checks meant to constrain it? Write to us at sean@thehumanpioneer.com.