To see how good AI is at scenario modelling, I asked it to predict the most likely outcome of the US-Israel-Iran conflict using all its knowledge. I also asked it to limit its prediction to one paragraph. A further paragraph was given to explain the prediction process. So here is the answer:
The most probable outcome—based on current expert analysis—is not total destruction, liberation, or a single ruling event, but a messy middle path: Iran is likely to emerge heavily damaged but still intact as a state, with its regime either surviving in weakened form or hardening into more overt military rule (e.g. IRGC dominance), while the U.S. achieves partial tactical goals but no decisive strategic victory. Airpower alone is widely assessed as insufficient to collapse the regime, meaning Iran will likely be “battered but not broken” and may even become more hardline internally . The conflict itself is most likely to de-escalate into a prolonged, lower-intensity regional struggle—proxy attacks, cyberwarfare, and intermittent strikes—rather than a 20-year Vietnam-style occupation, because neither side has strong incentives or capacity for full-scale ground war. At the same time, while asymmetric retaliation (including terror-style attacks) is a real risk, large-scale catastrophic attacks inside the U.S. remain possible but lower probability compared to regional escalation. Overall, the highest-probability scenario is: a weakened but surviving Iran, no clean regime change, continued instability across the Middle East, and a long tail of tension rather than a decisive end.
I didn’t run a full hidden war-game or a formal Monte Carlo model in the strict technical sense. I made a fast strategic forecast by combining a few high-weight factors: how wars like this usually end, what airpower can and cannot do, the resilience patterns of centralized authoritarian states under bombardment, the relative incentives of the U.S. and Iran, and the difference between tactical damage and strategic collapse. The core logic was: regime destruction is hard without invasion; invasion is extremely costly and therefore less likely; “liberation” from outside military pressure rarely arrives cleanly or quickly; Iran has strong incentives to retaliate asymmetrically rather than fight conventionally to the finish; and the U.S. usually prefers limited, deniable, or standoff pressure over open-ended occupation when the costs of escalation are high. So I mentally compared several scenarios—rapid regime collapse, long occupation-style war, contained punitive campaign, regional proxy escalation, major mass-casualty retaliation on U.S. soil—and then ranked them by plausibility. The “weakened but surviving Iran, followed by prolonged instability” scenario came out highest because it requires the fewest unlikely assumptions. I did it quickly because this kind of analysis is basically pattern matching plus causal weighting: once the key variables are identified, you can eliminate lower-probability paths fast and focus on the scenario that best fits how states actually behave under pressure.
I will check back in 12 months how accurate this prediction was.