Beyond the hype: What AI actually did in the latest US–Iran confrontation

The claims—and the record

A burst of Chinese-language posts claimed large language models steered U.S. “decapitation” strikes. A sober analysis by Huxiu (虎嗅) argues the truth is narrower—and more interesting. According to the Wall Street Journal, and echoed by other Western outlets, Anthropic’s Claude was integrated into U.S. Central Command (CENTCOM) intelligence workflows for tasks such as triaging data, identifying targets, and running simulations. Human operators remained firmly in the loop. It has been reported that the White House moved to restrict Claude’s use on security grounds, but that existing integrations made an immediate cutoff impractical—underscoring how quickly commercial AI has penetrated defense-adjacent analysis.

Grok’s “prediction” and the mythmaking

What about xAI’s Grok? Reportedly, a user posted prompts showing only Grok ventured a specific date—February 28—for a U.S. move, which later appeared to align with events. Impressive? Perhaps. Decisive? No. There is no evidence Grok was used by the Pentagon for targeting or intelligence; any suggestion otherwise collapses correlation into causation. It has also been reported that xAI’s outreach to replace Anthropic in defense contexts came late, making operational substitution implausible on the eve of action. A smart public-info guess is not the same as a classified AI workflow.

Drones aren’t generative AI

Some social media narratives folded advanced drones and autonomous navigation into a story of “AI-led warfare.” That stretches the term. Generative AI refers to models like GPT, Gemini, Claude, and Grok; long-standing “core AI” and automation—fire-control computers, cruise-missile guidance, flight autopilots—pre-date today’s LLMs by decades. Crucially, there is no evidence that LLMs executed commands in the kill chain. Claims that Palantir played a central role are largely traced to Reddit chatter; reportedly, no serious outlet has verified them.

What changed—and what hasn’t

So did AI matter? Yes, in the analyst’s chair. LLMs can compress messy, multimodal feeds—text, imagery, signals—into usable options, highlight patterns, and surface blind spots faster than humans alone. That’s why militaries are experimenting, even as they keep humans accountable for life-or-death calls. The open question is how much of Claude’s contribution here was mere efficiency (faster summarization) versus discovery (novel patterns and actionable insights). WSJ reportedly cites behavioral pattern analysis of Iranian leadership, but specifics remain opaque. As Washington tightens rules on commercial AI in sensitive workflows and allies weigh data-sovereignty risks, the geopolitical subtext is clear: AI is becoming a battlefield enabler—but not the battlefield commander.