US military command in the Middle East used Claude, the chatbot made by Anthropic, while preparing strikes on Iran, according to The Wall Street Journal. Sources said the system analyzed intelligence, helped simulate combat scenarios, and took part in planning missile attacks on Iranian targets. It remains unclear whether the model was used after the operation began.
This happened even though relations between the Pentagon and Anthropic had already deteriorated. After Claude was reportedly used in an operation against Venezuelan president Nicolas Maduro, the company said its internal rules prohibit using the product for weapons development, intelligence work, or violent actions. On February 28, 2026, US Defense Secretary Pete Hegseth accused Anthropic of arrogance and betrayal, while Donald Trump publicly attacked the company and barred federal agencies from using its service.
Yet the break was incomplete. In July 2025, Anthropic had signed a $200 million agreement with the Pentagon. Sources cited by Defence One said military specialists remain heavily dependent on Claude and cannot switch quickly to other systems; replacing it would take at least several months.
Against that backdrop, the Pentagon is already deepening cooperation with OpenAI. The company has begun integrating its models into a closed military network. OpenAI chief Sam Altman said two limits apply to work with the Pentagon: no mass surveillance inside the United States and no use of AI to control weapons of mass destruction.
Google, Microsoft, and Elon Musk’s xAI are also involved in US military AI projects. In the view of The Washington Post, major tech firms are entering such partnerships not only for contracts, but also out of concern over political and financial consequences if they refuse.
At the same time, military AI is being developed by China, North Korea, South Korea, Japan, India, Israel, and the EU. Experts warn that neural networks can speed up data analysis and improve military effectiveness, but without human oversight they may sharply increase the risk of mistakes and escalation.