Pentagon tests OpenAI and Google models

The Pentagon has begun testing artificial intelligence models from OpenAI, Google and xAI as possible replacements for Anthropic’s Claude inside military workflows, according to reports published on May 21 and May 22. Bloomberg reported that 25 Defense Department “power users” are evaluating rival systems to see which they prefer as the U.S. military looks for alternatives to Claude. The testing follows a breakdown in the Pentagon’s relationship with Anthropic, which had been the only AI company deployed on the department’s classified networks, CBS News reported on March 5. The Pentagon formally designated Anthropic a “supply chain risk” after a dispute over whether Claude could be used for “all lawful purposes,” including uses Anthropic said should remain off limits. (bloomberg.com) The immediate question is not only which model wins the comparison, but what exactly the Pentagon is trying to replace. Claude had already been embedded in defense workflows, which means the department is not merely buying new software; it is testing whether other frontier models can slot into existing tasks with limited disruption. Bloomberg’s account, as summarized in follow-on reports, said the review is focused on practical substitution rather than a greenfield procurement. (cbsnews.com) The Pentagon’s wider vendor lineup gives those tests a ready-made pool of alternatives. On May 1, the Defense Department announced agreements with eight companies — OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, Oracle, SpaceX and Reflection — to deploy AI capabilities on classified networks for “lawful operational use.” DefenseScoop, Breaking Defense and the department’s own release said those systems are being brought into Impact Level 6 and Impact Level 7 environments, the department’s classified cloud tiers. (bloomberg.com) That matters because xAI appears in Pentagon reporting through Elon Musk’s SpaceX, which is one of the companies named in the May 1 classified-network agreements. Reports tied to the current testing say xAI’s models are among the systems being considered alongside OpenAI and Google. Based on those two sets of reports together, the Pentagon is drawing replacement candidates from vendors it has already moved closer to on classified infrastructure. (defensescoop.com) Anthropic’s side of the dispute remains active. Dario Amodei, Anthropic’s chief executive, said on March 5 that the company did not believe the Pentagon’s action was legally sound and would challenge it in court. CBS reported that Anthropic has argued Claude should not be used for mass surveillance of Americans or for fully autonomous weapons, while Pentagon officials said existing law already bars unlawful conduct and that the department needed access for “all lawful purposes.” (thehill.com) Emil Michael, the Pentagon’s under secretary for research and engineering, said in March that OpenAI had already been deployed and that Google’s Gemini would follow, according to Breaking Defense. By May 1, the Pentagon had formalized classified AI agreements with a broader group of vendors, and by May 21 Bloomberg reported active user testing of replacements for Claude. (cbsnews.com) For companies building on a single model provider, the Pentagon episode offers a concrete procurement lesson. The department is testing whether existing workflows built around one frontier model can be shifted to rivals on short notice. That does not prove switching is easy in every enterprise setting, but it shows a large customer is spending time now on making that option real. (bloomberg.com) (breakingdefense.com)