The US and Israel struck Iran. The Supreme Leader is dead. Missiles are flying across the Gulf. Iran has closed the Strait of Hormuz, the artery for roughly 20% of the world’s oil supply.
This is a war, four days old, with no ceiling in sight.
Your AI systems didn’t get the memo. They’re still forecasting, still optimising, still producing clean confident outputs built on assumptions from a world that effectively ended last Saturday. Nobody reprogrammed them over the weekend. Nobody flagged the change. They just kept going.
The Houthi disruptions last year already showed how badly these models buckle under the kind of pressure nobody planned for. The Strait of Hormuz closure is a completely different scale of problem. Energy prices, freight costs, insurance premiums, currency exposure across emerging markets are all moving fast right now, in ways most models have genuinely never encountered. And here’s the thing about that: the models won’t surface it. They’ll keep printing outputs that look exactly as authoritative as they always did.
Iran has also spent years building cyber capability for precisely this kind of moment. Retaliation against Western and Gulf infrastructure isn’t a threat being drafted somewhere, it has already begun. If your systems have any connection to Gulf infrastructure, regional vendors, or networks now under active stress, that’s a conversation that needed to happen yesterday.
The companies that built real AI governance, not the annual review document but the actual living practice of knowing what their systems are doing and being able to intervene, are walking into this week with some degree of control. Everyone else is essentially running on hope and calling it operations.
One decision, made in Washington on a Saturday morning, dismantled the core assumptions underneath a significant chunk of global supply chains in under 48 hours.
I’ve said my piece. Now I genuinely want to hear from people who see this differently.
Is the connection between geopolitical conflict and AI governance risk being overstated here? Does your organisation actually have the mechanisms in place to respond to something like this in real time? Or are most of us quietly hoping the models hold together long enough for things to calm down?
Say what you actually think. Pushback is more useful here than agreement.
