US Central Command has used Anthropic's Claude chatbot as a key AI tool to help plan military strikes on Iran, including assessing intelligence, running simulated war games and identifying military targets, according to the Wall Street Journal. The joint US-Israeli campaign has killed 555 Iranians, including 165 in an attack on an elementary school in southern Iran.
The reported role of Claude in target selection sits awkwardly beside Anthropic's public positioning on military AI. The company and its CEO, Dario Amodei, have been in conflict with the Trump administration over two specific redlines: using Claude for surveillance of US citizens and for fully autonomous lethal weaponry. The Pentagon, which uses Claude across its classified systems, set a deadline last week for Anthropic to drop both redlines. Anthropic let the deadline pass without caving, a stance many observers read as a principled refusal.
What that refusal did not cover, according to Pulitzer Prize-winning national security journalist Spencer Ackerman, is the use of Claude to select targets in an active war. "Amodei, it is highly conspicuous, doesn't register building a surveillance panopticon of foreigners as a problem," Ackerman wrote. "The time to worry about everything ostensibly concerning Amodei was before signing the contract that Amodei didn't wish to abandon."
Where the redlines stop
Ackerman framed the arrangement bluntly: "When you take Doctor Doom's money to provide him a lathe to construct components for anthropomorphic robots, do you not understand that he is going to build Doombots?" The point is that the ethical guardrails Anthropic publicly defended did not extend to target identification, even as the strikes continued to claim civilian lives, including at a school.
Anthropic has separately dropped a major safety pledge that was described as central to the company's identity. The company has not publicly detailed how Claude's outputs are reviewed before military action, and the Wall Street Journal report did not specify the degree of human oversight in target selection.
Editorial Notes