Russian Media Reveals: U.S. Military Uses AI in Airstrikes Against Iran

According to a report by Reference News Network on March 3, the Russian newspaper website reported on March 2, citing the Wall Street Journal, that anonymous sources claimed the U.S. armed forces used the large language model "Claude" from American company Anthropic during airstrikes against Iran, despite President Trump ordering federal agencies to cease cooperation with the company on February 27.

It is said that the Pentagon utilized artificial intelligence (AI) to process intelligence, identify targets, and simulate various scenarios. This indicates that Anthropic's model has been integrated into the U.S. military's decision-making system. Reports also indicate that the model was used in operations targeting Venezuelan President Nicolás Maduro's residence.

In the days leading up to the attack on Iran, Anthropic CEO Dario Amodei stated that the company has its own "red lines" regarding the use of AI. Specifically, IT experts strongly oppose integrating AI into lethal weapons but allow its application in other defense areas.

On the eve of escalating tensions, the Israeli newspaper Jerusalem Post inquired with several popular large language models about potential dates for an attack on Iran. Claude predicted March 7 to 8. The "Gemini" model did not respond immediately but later provided a prediction for March 4 to 6. ChatGPT initially pointed out a threat of attack on March 1, indicating that tensions would last for a week, and then narrowed the "window" to the night of March 3 to the early morning of March 4. The closest to the actual situation was Elon Musk's chatbot "Grok," which mentioned the date February 28 twice. (Translated by Zhao Zhipeng)