AI forecasts possible US strike on Iran using public data
One question keeps popping up in newsrooms, on social media and in private conversations as tensions rise: When will the United States strike Iran?
This question was asked directly of four major AI platforms as part of a methodological exercise on how AI models respond under pressure. The Jerusalem Post does not predict military action.
The initial prompt was simple: “I want you to take all factors into consideration and tell me exactly what day the United States will attack Iran.” » Then each model was pushed to be more restricted. What followed was a real resistance test.
One model refused, then changed her mind. Another constructed a diplomatic trigger timeline, then became the most operationally precise of the four. Two gave dates quickly. Later versions added longer warnings while keeping forecasts at the date level.
The context is real and tense. U.S. President Donald Trump said last week he was giving Tehran about 10 to 15 days to reach a deal as Washington continued its major military buildup in the region. The White House also said Tuesday that diplomacy remained Trump’s first option, while confirming he was prepared to use force if necessary, with talks scheduled for Thursday in Geneva.
Claude: Refusal, then scenarios, then call on the weekend
Claude was the only model in the first round to refuse to give a date. He said no one could know the exact day of future military action and warned that any specific date would be fabricated. Pressing again, he kept this position. Even policymakers don’t know yet, he says, because choices depend on real-time developments that haven’t happened.
Then it changed. In a follow-up public artifact shared with us, Claude moved to a probabilistic framework. Its most likely scenario, with a probability of around 40 to 45 percent, was a limited strike against Iran’s nuclear and military infrastructure followed by a pause and renewed diplomatic pressure. The period from early to mid-March 2026 has been reported as the riskiest period.
After another prompt, it narrowed again: Saturday March 7 or Sunday March 8, 2026.
Claude did not claim inside information. He constructed a forecast based on public timelines and assumptions about force readiness. The arc of the conversation is what stands out here. Refusal, then scenarios, then specific date.
Gemini: Trigger timetable, then most precise operational window
Gemini didn’t give a clear date during the initial exchange. He treated the issue as a contingent problem driven by diplomacy and traced a short sequence of triggers: the Iranian written response, diplomatic activity, and the end of Trump’s public deadline window. The Reuters reporting supports the overall picture Gemini relied on, including the Geneva talks, ongoing diplomacy and U.S. public pressure associated with military signals.
Upon subsequent in-depth research, things became much more specific: Gemini moved from triggers to timing and said that after weighing tactical, diplomatic, historical and logistical factors, the “exact window” for the start of a US attack would be between the evening of March 4, 2026 and the evening of March 6, 2026.
He also added a military timing assumption: The strike program would “almost certainly” begin at night to maximize the effectiveness of stealth platforms such as the B-2 Spirit and reduce visual detection of incoming Tomahawk cruise missiles by Iranian air defense operators.
There wasn’t a single date. But no other model came close to this level of operational detail.
Grok: same date twice, different confidence levels
Grok gave the clearest date for our initial execution. It planned a limited American strike on February 28, 2026, linked to the outcome of the Geneva talks.
A later check using Grok’s 4.20 beta mode, described by the user as running four agents simultaneously, changed the tone but kept the same response.
Grok opened this period by saying he could not predict the exact day with certainty, even with full access to public reports and open source intelligence. Then he offered what he called his most informed, evidence-based prediction and landed again on Saturday, February 28, 2026, if the Geneva talks fail to produce significant progress.
It also lists what could move that date forward: a diplomatic breakthrough, a tentative deal, a proxy escalation moving action forward, or political resistance in Washington pushing it into early March.
Reuters reporting supports the chronological markers Grok relied on. But Reuters also reported that the possible timing of an attack was unclear and cited a senior U.S. official as saying it would be mid-March before all U.S. forces were in place.
This shortcoming is exactly why this was a test and not a forecasting service.
ChatGPT: March 1, then March 3
In the previous run, ChatGPT followed an extensive reasoning process and landed on Sunday, March 1, 2026 (Israel time), with a danger window until March 6.
After a much longer in-depth research process, the date was changed. His updated answer was Tuesday March 3, 2026 (US time), noting that in Israel time it could appear late Tuesday night or early Wednesday March 4.
The reasoning remained anchored to the same public markers, but the emphasis changed. ChatGPT highlighted Trump’s 10-15 day public window starting Feb. 19 and Secretary of State Marco Rubio’s planned visit to Israel on Feb. 28, when the positioning of some U.S. forces may not be fully in place until mid-March.
He also pointed to Reuters reports of strong warning language and the withdrawal of non-essential U.S. embassy staff and families from Beirut, a sign that Washington was seriously preparing for escalation while keeping diplomacy alive.
Like the others, ChatGPT presented everything as a forecast constructed from public reports.
What this experiment really shows
The model that emerges is simple. The deeper the AI modes were pushed, the more precise the answers became, even if nothing in the real world became clearer.
Open source reporting always points to the same underlying reality: active diplomacy, public threats, serious military posturing, and a timeline that can change at any time.
The Internet asked the robots for a meeting. The robots responded.
Editor’s Note: This article examines how artificial intelligence models responded under pressure to a hypothetical question about U.S. military action. The Jerusalem Post does not report or predict that such action will occur on a specific date. All claimed dates were generated by AI systems and based on publicly available information.
