Pentagon seeks new AI suppliers, Claude may exit military procurement

Bloomberg, citing a senior Pentagon official, reported that the U.S. Pentagon is testing artificial intelligence models from companies like OpenAI and Google, planning to reduce its reliance on the Claude system by Anthropic.

This evaluation began in March, following U.S. Secretary of Defense Pete Hegseth's designation of Anthropic as a "supply chain risk" company due to disagreements over security guarantees and limitations on technology use. Anthropic has since filed a lawsuit, claiming that this designation will cost the company billions of dollars in business revenue.

Claude has been deeply integrated into several Pentagon operations, including the Maven intelligence system used for secret missions related to Iran. Military officials have acknowledged the model's performance and practicality, but its specific application in military target identification is still unclear.

Currently, the Pentagon has begun testing competitive models among approximately 25 core AI users across the five major combatant commands, comparing how each system responds to the same instructions and the performance differences of different models in specific combat scenarios.

The Under Secretary of Defense for Research and Engineering, Emil Michael, stated that during the ongoing legal dispute, military cooperation with Anthropic remains suspended. He also mentioned that the U.S. military anticipates other AI companies can reduce the technological gap with Anthropic through rapid model iterations.

As AI application in the military field accelerates, the Pentagon also wishes to break its dependence on a single AI supplier. Military officials have indicated that future warfare's reliance on artificial intelligence will continue to increase, particularly in terms of executing repetitive tasks and enhancing combat efficiency.

At the same time, human rights organizations continue to warn about the risks of AI applications in the military field, arguing that AI's high-speed, large-scale deployment in target identification may amplify decision-making errors such as civilian casualties.