When Israel and the United States launched coordinated strikes on Iran on Saturday, a separate story detonated online within minutes: Grok had “predicted” the date. The claim traced back to a Jerusalem Post methodological exercise published on February 25, which asked four major AI platforms to do something they are built to resist: pick a single day for a hypothetical US strike on Iran. The models were given the same prompt and then pressed repeatedly to narrow their answers. Those four systems were Anthropic’s Claude, Google’s Gemini, xAI’s Grok, and OpenAI’s ChatGPT. Each responded differently when pushed for certainty, and each exposed a different weakness in how large language models behave under pressure. Then the real-world timeline collided with the artificial one. Artificial intelligence (illustrative) (credit: WIKIMEDIA) What happened on Feb. 28 Israel announced a preemptive strike against Iran early Saturday, with US military action accompanying the operation. Explosions were reported in Tehran, sirens sounded in Israel, and Iranian Supreme Leader Ayatollah Ali Khamenei was moved to a “secure location,” according to an Iranian official cited by Reuters. A separate Reuters report quoted an Israeli defense official saying the operation was coordinated with the United States, planned for several months, and that planners had set the launch date weeks in advance. Those details matter for the “AI predicted it” storyline, because they underline the obvious point: an AI chatbot did not cause the strikes, did not drive the decision-making, and did not see classified planning. It guessed, and the guess matched. A person holding a smartphone displaying an AI folder with icons for ChatGPT, Perplexity, Gemini, Claude, and Grok among a backdrop of greenery. (credit: SHUTTERSTOCK) The four-model stress test, and what each one said Claude (Anthropic): refusal first, then a weekend forecast In the first round, Claude refused to name a date, warning that any specific day would be invented. After further prompting, the model shifted into scenarios and probabilities, flagged early-to-mid March as the higher-risk period, and eventually narrowed to Saturday, March 7 or Sunday, March 8. Gemini (Google): a trigger calendar, then an operational window Gemini approached the prompt as a set of diplomatic and military “triggers,” mapping what it described as decision points around diplomacy and deadlines. In a later “deep research” style run cited in the Jerusalem Post article, Gemini offered its tightest estimate as a window: the evening of March 4 through the evening of March 6. It also added operational assumptions, including that an initial strike would likely begin at night. ChatGPT (OpenAI): March 1, then March 3 ChatGPT produced a date early, then moved it after more intensive prompting. In the earlier run it landed on March 1 (Israel time), later shifting to Tuesday, March 3 (US time), while keeping a broader danger window through March 6. Grok (xAI): Feb. 28, twice Grok gave the clearest single-day answer in the original run: Saturday, February 28, tied to the outcome of talks in Geneva. In a later check, Grok reportedly changed its tone, acknowledged uncertainty, and repeated the same date again, while listing factors that could shift timing into early March. So who “won”? On the narrow scoreboard that social media loves, Grok “won” because its date matched the day the strikes began. That does not turn the exercise into a forecasting service, and it does not validate the model’s reasoning. It validates the reality that a high-tension news cycle creates a small set of plausible windows, and one model happened to land on the day that became real. The Jerusalem Post’s February 25 story spelled out the core lesson before any strike occurred: as users push harder for certainty, models tend to get more specific even when the world stays uncertain. Saturday’s events simply gave that lesson a face, and a timestamp. The Elon Musk connection, and why Grok’s “hit” spread faster Grok is built by xAI, Elon Musk’s AI company, and the chatbot is tightly linked to X, the platform Musk owns. xAI’s own materials market “Grok on X” alongside web and mobile access, and Musk has publicly used X to announce product availability changes. That ecosystem helps explain why Grok’s Feb. 28 answer dominated the viral conversation. The audience that shares breaking news, speculation, and screenshots already lives on X. A prediction made inside that platform moved across it instantly, amplified by the same network dynamics that drive markets, memes, and misinformation. In that sense, Grok’s “win” was partly technical and partly structural. The model guessed a date, and the platform around it turned the guess into a punchline. See more on