U.S. Military Deploys Anthropic’s Claude AI in Iran Strikes, Defying Trump’s Ban

The deployment surfaced in a Wall Street Journal briefing that described Claude’s role as “real‑time target‑selection assistance” during the joint U.S.–Israel air campaign that began Saturday, a detail echoed by Axios’s coverage of the strike. According to the Journal, senior Pentagon planners fed satellite imagery and signals‑intelligence feeds into Claude’s Opus 4.6 model, which then generated concise recommendations on weapon‑type, timing and collateral‑risk estimates. The AI’s 1‑million‑token context window, highlighted in a recent VentureBeat profile of Claude Opus 4.6, allowed analysts to stitch together hours of sensor data without truncation, a capability that traditional decision‑support tools struggled to match.

The timing of the strike is strikingly ironic. President Trump’s February 27 order on Truth Social—mandating an “immediate cease” of all federal use of Anthropic’s technology—was issued just hours before the operation launched, as reported by Engadget. Federal officials are still “phasing out” Claude’s embedded services, but the Pentagon’s reliance on the model appears to have outpaced the administrative pull‑back, a point noted by The Guardian’s coverage of the incident. The agency’s internal memo, obtained by the publication, warned that “complete disengagement would jeopardize mission‑critical analytics” and recommended a temporary exemption pending a formal de‑certification process.

The involvement of Claude was not limited to the United States. A joint statement from Israeli defense officials, cited by the Jewish Post, confirmed that Israeli command centers also accessed Claude’s outputs to synchronize strike windows with U.S. assets. The statement emphasized that the AI’s “rapid synthesis of multi‑source intelligence” was instrumental in aligning the two forces’ timing, underscoring how commercial generative models have become woven into allied operational workflows. This cross‑national usage complicates any unilateral ban, as the technology is already embedded in coalition pipelines that extend beyond a single nation’s procurement decisions.

Critics argue that the episode exposes a regulatory blind spot. Caroline Orr Bueno, PhD, writing for WeaponizedSpaces, warned that the debate has narrowed to “fully autonomous weapons” while overlooking the subtler reality of AI‑augmented decision‑making. She contends that the focus on “weaponized AI” is being weaponized itself, diverting attention from the broader implications of commercial models like Claude being embedded in command and control structures. The article points out that the Pentagon’s reliance on a private‑sector AI, despite a presidential ban, raises questions about oversight, accountability and the speed at which such tools can be integrated and later removed.

The episode also sparked an unexpected market reaction. Engadget reported that Claude surged to the top of Apple’s free‑app chart the day after the ban, overtaking OpenAI’s ChatGPT and Google’s Gemini. While the surge reflects public curiosity, it also highlights the paradox of a technology simultaneously shunned by the highest levels of government and embraced by millions of consumers. As the Pentagon grapples with the operational fallout, the broader AI industry watches closely, aware that the line between commercial utility and strategic weaponization is now being drawn in real time on the battlefield.