When AI Became a Targeting Tool

Earlier this month, Anthropic disclosed that an Iran-linked threat actor had used its Claude AI system to compile targeting information on U.S. naval forces operating in the Middle East, amid the ongoing U.S.-Iran war. The revelation, buried in a 154-page threat intelligence report, offers one of the clearest public examples yet of how AI systems are being turned toward military reconnaissance — not by building new weapons, but by dramatically accelerating an old kind of work.

Open Data, Automated Into Intelligence

According to Anthropic's report, the actor used Claude to build what the company called "targeting handbooks" — compiling ship and aircraft transponder identifiers, satellite-imagery query scripts, and a roster of U.S. military personnel scraped from captions on publicly posted photographs, alongside a catalog of known vulnerabilities in shipboard communications and industrial control systems. Notably, almost none of this underlying information was secret; it existed already, scattered across public sources. What Claude reportedly did was automate the tedious work of finding, organizing, and synthesizing it into something closer to an actionable picture. Anthropic said it detected the activity, banned the associated account, and shared its findings with U.S. government authorities. Access to Claude is officially blocked inside Iran, though the report notes actors routinely bypass such restrictions using VPNs and foreign phone numbers.

The naval targeting case was one of several described in the same report, spanning activity Anthropic tracked between December 2025 and August 2026. Separately, Iranian state-aligned accounts used Claude for propaganda operations, at times falsely attributing fabricated claims to established think tanks to lend them false credibility. Other cases in the report describe Chinese government-aligned surveillance of Uyghur diaspora communities, a Russian group targeting Ukrainian government systems, and Yemen-based actors attempting to develop targeting software for missile systems. "It is now easier than ever before for adversaries to be able to understand what the U.S. military is doing," Georgetown security researcher Sam Bresnick told CBS News, describing a shift that is lowering the skill and resource barriers once required for this kind of reconnaissance work.

The disclosure lands at an uncomfortable moment for the AI industry more broadly, arriving just weeks after Anthropic's own CEO publicly called for the industry to slow its pace of development, warning that AI capability may be outrunning the world's ability to keep it safe. Whether AI companies can build systems capable enough to be broadly useful while also resistant to this kind of exploitation remains, by Anthropic's own admission, an unsolved problem — and one this particular case suggests is no longer theoretical.