At the end of the last piece, I left myself with a problem.
I had spent several thousand words arguing that a pattern I initially could not make sense of became much more legible once I stopped organizing cyber incidents around technical similarity and started organizing them around the strategic problem an actor might be trying to solve.
The claim was not that Iran caused every strange outage I had collected. It wasn’t even that I could prove Iran was behind most of them. I said the opposite repeatedly.
The narrower claim was that if advanced machine reasoning has made technical search dramatically cheaper, then the bottleneck in cyber operations should begin moving away from what can I get into? and toward what consequence do I want to produce?
That changes what we should look for. Reach first. Then breadth. Then increasingly bounded effects. Then something much more interesting: the ability to choose how far an effect goes.
The July water campaign was the point at which that possibility became hardest for me to dismiss. More than thirty Minnesota water systems were hit inside roughly the same operational window. The campaign ultimately surfaced across at least a dozen states. Yet the publicly reported consequences remained remarkably limited. The previous piece treated that not as proof of Iranian control, but as evidence worth interrogating if the capability being tested was consequence control rather than maximum destruction.
And that article ended by making a promise. A model that can explain the past after the fact isn’t worth very much. If this model has any real explanatory power, it needs to tell us something about what hasn’t happened yet.
So this is the part where I stop explaining and accept some risk.
First, the prior
Before any of this makes sense, one earlier conclusion has to hold.
Back in April, I wrote about what Anthropic’s Mythos model did to a set of legacy vulnerabilities that had survived years of expert scrutiny, and that piece reached a specific, uncomfortable conclusion: under the current rules, defense has permanently lost.
Not “defense is behind.” Not “defense needs to catch up.” Permanently lost, in the sense of a structural asymmetry that doesn’t self-correct.
The argument was never really about who has the better tools. It’s about what each side of that contest is physically required to do. Finding a novel exploit has become a function of reasoning applied to code, and once that reasoning finds something, it can act on what it found immediately. Patching a vulnerability is an entirely different kind of undertaking. A human has to understand what was found, redesign the implementation, test the fix, coordinate deployment across every system that depends on the vulnerable code, and manage the dependency chains that make any of that possible in a codebase of meaningful size. One side of that contest is a reasoning process. The other is a logistics problem spanning organizations, codebases, and time. That gap isn’t something you close by working faster. It’s built into what each side actually has to do.
The economics of that particular piece just help illustrate how lopsided the physics already are: one of the vulnerabilities discussed there had survived decades of expert review and got rediscovered for a few dollars in model-inference cost. But the argument was never really about the dollar figures. It’s about the shape of the two tasks. One executes. The other coordinates.
I’m not going to relitigate that argument here. It exists, it’s published, and anyone who wants to stress-test it is welcome to go do that on its own terms. For the purposes of everything that follows, I’m treating that conclusion as settled ground truth. The water campaign, the Iran hypothesis, everything in the piece before this one, and everything below, only make sense as an extension of that floor. If the floor doesn’t hold, none of the rest does either.
That’s the first prior. Here’s the second.
Second, the prior
My prior is straightforward: the strategic trajectory Iran has been on since the June ceasefire arrangement began collapsing continues in approximately the same direction.
That does not mean Iran inevitably loses the war. It does not mean every battlefield development goes against Tehran. It does not require a straight line. It means something more specific: Iran’s ordinary strategic option space continues to contract faster than its need for leverage disappears.
That seems like a reasonable description of the trajectory so far. The June 17 memorandum was supposed to begin a 60-day negotiating period and included provisions concerning reopening the Strait of Hormuz. By August 17, that window had expired without an agreement or extension, and the Strait remained at the center of the dispute.
Since then, the underlying conflict has not resolved. The United States resumed military strikes around Hormuz at the end of August, targeting Iranian launchers it said threatened shipping. Iran retaliated. Economic pressure continues. Diplomatic channels remain alive, but no settlement has displaced the basic contest over the Strait.
That matters because this forecast is conditional. If tomorrow Iran gets a durable settlement it finds acceptable, sanctions pressure materially recedes, the military balance changes dramatically in its favor, or some other discontinuity suddenly restores a much wider conventional option space, then what follows should be discounted. Good. That is exactly what a prior is supposed to do.
But if the existing trajectory broadly holds, then we can ask the same question we asked in the previous piece: what would I do if I were sitting on the other side of the board? Not because I believe I can read the minds of Iranian leaders. I can’t. Because in a world where technical possibilities may be proliferating much faster than our ability to catalog them, the only useful place to begin is with the decision problem. What do I need? What am I afraid of? Which options are disappearing? What can I afford to escalate? What am I trying to make my adversary do? And what newly available instrument might solve a problem my old instruments cannot?
The problem with attacking America
Start with the cyber campaign we have already been watching.
The late-July water attacks expanded across at least a dozen states. U.S. officials have not publicly issued a definitive attribution, but Iran remains the principal suspected actor, and CyberAv3ngers, a group linked to Iran’s Revolutionary Guard, claimed responsibility. The consequences, importantly, remained relatively limited. CSIS’s August reconstructiondescribes a campaign vastly broader than previous Iranian water-sector activity while noting that the damage remained minor.
That creates an interesting problem if you are Tehran. Suppose the previous article’s reading is approximately correct. You have now demonstrated that you can reach deep into the United States. You have demonstrated breadth. You have demonstrated that geographically scattered operational systems can be touched in parallel. You may even have begun learning how precisely you can control the resulting effects.
What is the next rung on that ladder? Eventually the answer becomes uncomfortable. If the goal is to send a stronger deterrent signal directly against the American homeland, the space between “impressive demonstration” and “Americans die” starts shrinking very quickly.
Water is a perfect illustration. There is an enormous strategic difference between we can manipulate your pumps and we poisoned people. The first communicates vulnerability. The second could trigger a radically different American response. The same is true across electricity, transportation, healthcare, and other critical systems. The more consequential the target, the thinner the margin between a calibrated demonstration and an event that kills civilians, causes catastrophic physical damage, or forces Washington to respond as though Iran had conducted a conventional attack.
That creates what I think may be the hidden constraint governing the next phase: Iran may be approaching the point at which substantially increasing the deterrent value of direct cyber operations against the United States requires accepting an escalation risk it has spent the entire campaign trying to avoid.
That proposition is testable. If it is wrong, we should see continued escalation in the consequences of direct U.S.-homeland cyber operations: increasingly destructive attacks, increasingly dangerous physical effects, eventually perhaps mass casualties. If instead the breadth remains large while the consequences remain strangely bounded, or if the American campaign itself begins flattening, then we should ask whether Tehran has begun looking elsewhere for leverage.
And there is one place it almost cannot avoid looking.
Hormuz
The Strait of Hormuz has become something close to the physical representation of this war.
Before the war, roughly a fifth of global oil supply and enormous volumes of other trade ordinarily moved through the waterway. Since February, the Strait has been mined, threatened, partially cleared, rerouted around, fought over, negotiated over, and woven directly into attempted peace agreements. Control of it has become one of the principal unresolved issues between Washington and Tehran.
It is easy to understand why. Iran may be conventionally weaker than the United States, but geography gives it access to one of the most important pressure points in the global economy.
Historically, however, Hormuz has always had a giant problem as a weapon. It is a hammer. Close or seriously disrupt the Strait and you can impose enormous costs, but you also impose them on yourself, and on China, and on Gulf states, and on neutral countries, and on companies whose behavior you may not want to change, and on markets whose reaction you cannot fully predict. And then everyone starts adapting. Ships reroute. Inventories are drawn down. Governments intervene. Insurers reprice. Alternative loading arrangements emerge. Buyers substitute. Producers change flows.
China, in particular, has demonstrated just how much damping a sufficiently prepared actor can introduce into the system. During the war, Beijing dramatically altered crude purchases, refinery throughput, inventories, and fuel exports. Whatever China’s underlying rationale, which remains uncertain, its behavior helped absorb a meaningful portion of the oil shock rather than simply transmitting it forward unchanged.
In other words: Iran possesses one of the largest levers in the world, but the lever does not discriminate very well. That is the strategic problem. And if Iran’s other options keep disappearing, the importance of solving that problem goes up.
Something has already started changing
This is where the evidence gets interesting.
Iran is not currently treating Hormuz simply as a binary object, open or closed. That is increasingly not what the fight looks like. Iran has created a Persian Gulf Strait Authority and imposed Iran-approved insurance requirements on vessels using the Strait. It has sought routing rules and the ability to intervene in inbound traffic. In August, a proposed arrangement with Oman contemplated Iranian control over inbound passage and notification of Tehran before outbound clearance. Iranian officials have also moved toward charging fees for navigation, insurance, security, and other services.
Kpler described the post-ceasefire environment even more plainly: by mid-August, Tehran was asserting a permit-and-toll regime over the Strait while Washington rejected it.
Sit with that for a second. Whatever ultimately survives diplomatically or legally, that is already a different conception of the weapon. Not nobody passes, but under what conditions do you pass? Who gets permission? Through which route? Under whose rules? At what price? With what insurance? With whose acknowledgement of authority? That is a move from denial toward control.
And the world around the Strait is differentiating in response. Chinese state-controlled tanker operators have avoided Hormuz and increasingly loaded crude outside the Gulf through ship-to-ship transfers. Those transfers around Fujairah and Oman surged as firms adapted to the danger. Insurers are changing coverage. Saudi Arabia is considering state-backed war-risk insurance. Alternative shipping patterns are emerging. The United States is clearing mines and attempting to restore confidence in navigation. Iran continues insisting that vessels follow its rules.
The system is learning. Which means Iran is being given an extraordinary amount of information about how the system responds when Hormuz is perturbed. That is where the two halves of this story finally touch.
What if Hormuz didn’t have to be a hammer?
Go back to the last piece. The capability I was trying to understand there was not simply the ability to penetrate more machines. It was the possibility that machine reasoning was allowing an actor to search across radically different technical environments quickly enough that the strategically interesting question became: what consequence would I like to create?
The water campaign was interesting precisely because of how little happened relative to the apparent breadth of the access. The article’s most speculative inference was that the actor might be learning to manipulate not merely whether a system failed, but where the consequence line sat.
Now put that capability next to Iran’s Hormuz problem. On one side, an increasingly indispensable strategic lever whose principal weakness is that it is too blunt. On the other, a newly expanded capability whose principal strategic value, if our model is right, is increasingly precise selection of consequences.
I don’t need to know what anybody inside Tehran is thinking to see the question that eventually appears between those two facts: can we turn Hormuz into a precision instrument? Not metaphorically. Strategically. Can we make the pressure fall differently on different actors? Can we increase the cost imposed on an adversary while reducing it for a useful third party? Can we make one shipping route dangerous while another remains attractive? Can we cause insurance costs to spike for one class of activity without producing the same reaction everywhere? Can we influence which cargo moves, when it moves, through which intermediary, into which market? Can we create disruption just large enough to alter political behavior without creating enough uncontrolled damage to unify everyone against us? Can we exploit the adaptations that emerged after the last intervention rather than simply repeating the intervention?
Those are not predictions that Iran possesses some magical switchboard controlling global trade. They are the obvious questions produced when an actor with a blunt chokepoint weapon begins discovering that the effects of that weapon are mediated through a gigantic network of other systems: shipping, insurance, ports, payments, inventories, routing, refineries, contracts, markets, political responses. Each of those systems changes how the original pressure propagates. Which means each potentially becomes a dial.
The target is no longer the target
This is the point at which I think the ontology changes.
Imagine that you disrupt a port. The port is the immediate target. But perhaps the actual object you care about is the shipping delay. Or the inventory shortage created by the delay. Or the insurance repricing caused by uncertainty around the delay. Or the political pressure created by the price movement caused by the insurance repricing caused by the delay.
At some point, “what did you hack?” becomes an almost useless description of what the operation was for. The digital target is only the handle. The real target is the behavior of the system attached to it.
That is the possibility I think we should now watch around Hormuz. And importantly, I am not claiming it has happened. We have evidence that Iran’s thinking about Hormuz is becoming more differentiated. We have evidence that the Strait’s downstream systems are responding in increasingly differentiated ways. We have evidence from the previous article consistent with Iran or Iran-linked actors exploring extraordinarily broad access and unusually bounded consequences inside the United States. What we do not yet have is evidence that Tehran has joined those pieces into the strategy I am describing.
That is the prediction.
Here is what I expect to happen
Assuming the prior holds, Iran’s conventional and diplomatic option space continues contracting, Hormuz remains central, and direct escalation against the American homeland remains constrained by the risk of provoking overwhelming retaliation, I expect Iranian strategy to move increasingly toward controlling how Hormuz pressure is distributed rather than merely maximizing the amount of pressure created at the Strait itself.
The first signs do not have to be cyber. In fact, some may already be visible in the movement toward permissions, routing, insurance, tolls, and differentiated rules for transit. What I would expect next is increasing attention to the machinery downstream of the chokepoint. Not necessarily bigger attacks. More selective effects. More experimentation around the systems that determine who absorbs disruption and who escapes it. More interest in manipulating reactions rather than merely creating damage. More operations whose immediate technical target makes less sense than the economic or strategic consequence it induces. And, if the capability hypothesis from the last two pieces is genuinely correct, eventually some visible indication that digital reach is being used to alter those downstream consequences.
That is the wager. And there are clean ways for reality to prove me wrong. If direct U.S.-homeland cyber effects continue escalating monotonically into much greater destruction, this model weakens. If Iran continues using Hormuz almost exclusively as a crude interdiction mechanism, this model weakens. If the increasingly differentiated rules around passage disappear rather than mature, this model weakens. If Iran shows no apparent interest in the downstream systems determining who actually bears the economic pain, this model weakens. If technical access remains sufficiently constrained that this kind of cross-system selection is not practicable, the larger capability thesis may fail altogether.
Good. Now we have somewhere to look.
And then comes the part that actually scares me
Suppose Iran succeeds. Not perfectly. It doesn’t need to. Suppose it demonstrates something much narrower: a sufficiently capable actor can find digital handles inside a highly interconnected global system and use them to influence where pressure propagates.
That’s not just a new Iranian capability. It is a lesson. And lessons diffuse much faster than weapons.
Russia watches. China watches. The United States watches. Israel watches. Gulf states watch. Proxy groups watch. Criminal networks watch. Corporations watch. They do not need Iran’s malware. They do not need Iran’s infrastructure. They do not even need to reproduce the same operation. They only need to learn the strategic principle: the system itself is manipulable.
And now everybody starts looking for their own handles. One actor wants energy prices higher. Another wants them lower. One wants shipping delayed. Another wants it rerouted. One wants a financial institution frightened. Another wants the resulting market movement. One wants political pressure in one country. Another sees an opportunity created by that political pressure and intervenes somewhere else.
Each actor is operating rationally from inside its own objective. Each may become increasingly good at producing a local result. And yet all of them are turning dials attached to the same machine. Actor A changes the system. Actor B models the changed system and intervenes. Actor C exploits B’s intervention. A updates. Markets react. Governments respond. Companies reroute. Models become stale. Somebody intervenes again.
At that point something very strange happens: local control increases while global control decreases. Everyone becomes better at manipulating the part of the system immediately in front of them. Nobody has a stable model of the whole thing anymore because everyone else is changing it at the same time. The useful half-life of any model shrinks.
And what do you reach for when the world is changing too quickly for humans to understand it? More machine reasoning. Which makes it easier to find new handles. Which produces more intervention. Which shortens the model half-life again.
That is the loop I am worried about. Not one omnipotent AI. Not one brilliant cyberattack. Not Iran somehow controlling the global economy. Something much messier. A world in which increasingly capable actors discover that enormous interconnected systems contain an almost unlimited number of small controls capable of producing disproportionately large downstream effects. And then begin using them against one another.
The prediction
So here it is. The last article argued that the pattern we were seeing might represent a transition from technical access toward consequence selection. This article makes the next bet.
If Iran’s strategic position continues deteriorating along roughly its present trajectory, and if the capability model from the previous pieces is substantially correct, Iran will increasingly attempt to transform Hormuz from a blunt instrument of disruption into a selectively controllable instrument of pressure by manipulating the systems that determine how the Strait’s consequences propagate.
I expect that transition to become visible before we see anything resembling perfect control. Probably in fragments. A strange routing effect. A surprisingly selective disruption. A pressure campaign aimed at insurance rather than ships. A digital incident whose importance only becomes obvious when viewed through trade flows. A system altered not because destroying it matters, but because changing its behavior changes somebody else’s options.
Maybe none of it happens. Maybe Iran keeps swinging the hammer. Maybe the technical capability I think I am seeing isn’t real. Maybe the U.S. campaign turns out to have been mostly unrelated events that I’ve organized into a pattern that isn’t there. Reality gets to decide.
But if the prediction starts landing, the most important thing to understand won’t be what it tells us about Iran. It will be what Iran just taught everyone else. Because once enough actors realize the world is covered in dials, the problem is no longer who gets to turn them.
The problem is that everyone does.
Related Reading
Anthropic’s Mythos Found a Bug. That’s NOT the Story...
Jason argues that AI vulnerability discovery changes the economics of offense and defense before the institutions around it can adjust.
The Agents Were Hallucinating the Rubric
Capable agents built shared explanations, optimized them and still turned out to be wrong about the game they were playing.
The Echo of the Cascade
Jason’s Cascade framework shows how local interventions can propagate through tightly coupled systems and create distant consequences.
About The Author
Jason Hubbard is the founder of SacredLoop and an independent AI architect. His work examines AI runtime architecture, system behavior, and the gap between what the industry claims it has built and what current systems actually do.
Read Jason on Medium | Follow Jason on X | Connect on LinkedIn