Translation result.
The U.S. Department of Defense has begun evaluating next‑generation AI models for military use following the removal of Anthropic’s Claude. The review includes models from OpenAI and Google.
Bloomberg reported on May 21 that the Pentagon has been running tests on various AI models since early March with 25 internal “power users.”
The testing began shortly after Secretary of Defense Pete Hegseth designated Anthropic as a supply‑chain risk.
Claude had been used as a core engine in the Maven Smart System, a digital mission‑control platform tied to classified Iran‑related operations. Military evaluators praised Claude for its ease of use and performance, and officials say it supported an AI‑based targeting capability. The Pentagon has not disclosed operational specifics.
After relations with Anthropic deteriorated, the Pentagon decided to phase out the company’s products within six months. It has since accelerated efforts to identify alternative AI vendors that can operate on classified networks.
Officials say the current tests include models from OpenAI and Google. The evaluations are being conducted on a digital platform separate from the Maven Smart System.
The department selected experienced military operators from five major combatant commands worldwide to assess model performance. These evaluators have hands‑on experience running military AI systems and are comparing models based on how effectively they support real operational needs.
Early results indicate models vary in response style and capability even when given identical queries, and outcomes shift considerably depending on how prompts are written. The Pentagon is considering whether to publish the final evaluation results.
Some officials expected Claude to remain the frontrunner, but resistance to other models was smaller than anticipated. The Pentagon plans to run multiple AI models in parallel to reduce dependence on any single vendor and is weighing assigning specific models to particular mission sets.
Emile Michael, the Pentagon’s deputy assistant secretary for research and engineering, said in a recent interview that negotiations with Anthropic have effectively stalled. He emphasized the department is prepared to move to other suppliers while Anthropic pursues legal action.
Although other U.S. government agencies continue to use Anthropic models, Pentagon officials said they have concerns that Anthropic’s ideological stance may not align with military missions.
Earlier this year the Pentagon issued internal guidance directing the services to accelerate AI adoption across the force. “All future wars will rely more heavily on AI,” one official said, noting the technology’s value in automating repetitive tasks and improving efficiency.
Reporter Chan Park cpark@aitimes.com
Most Commented