Pentagon tests OpenAI and Google
- The Pentagon began operational testing of OpenAI and Google AI models this week as it looked for alternatives after restricting Anthropic systems. - Bloomberg reported 25 Pentagon “power users” are evaluating rival models, while xAI already holds a Pentagon contract worth up to $200 million. - Senators pressed Pentagon officials on May 20 over a proposed $54.6 billion autonomous-warfare budget and the doctrine governing its use.
The Pentagon has begun testing OpenAI and Google artificial intelligence models inside the Defense Department as it searches for alternatives to Anthropic’s Claude, according to Bloomberg and other reports. The review is being run with 25 of the department’s “power users,” Bloomberg reported on May 21, citing a senior defense official. The testing comes after the Pentagon restricted Anthropic systems on supply-chain grounds, widening a scramble among commercial AI firms to secure defense work. The shift is happening alongside broader military spending on autonomy. The Pentagon is seeking $54.6 billion for its Defense Autonomous Warfare Group, or DAWG, in the fiscal 2027 request, according to Defense One and Breaking Defense. Senators on the Senate Armed Services subcommittee on emerging threats said this week that policy for autonomous weapons may not be keeping pace with the scale of the spending. (bloomberg.com) ### Why is the Pentagon testing OpenAI and Google now? Anthropic was designated a supply-chain risk earlier this year, and Pentagon AI chief Cameron Stanley said in April that relying on one model provider was “never a good thing,” according to CNBC. Bloomberg reported the Pentagon is now comparing alternatives to see which systems are preferred by heavy internal users. The immediate effect is a broader vendor field for defense AI work rather than a single replacement. (defenseone.com) Google’s Gemini models are already part of that expansion. CNBC reported on April 28 that the Defense Department had increased its use of Google after the Anthropic restriction. Bloomberg’s May 21 report added that OpenAI models are also in the current operational testing group. ### What exactly is being evaluated inside the department? (cnbc.com) Bloomberg said the review is focused on which models are most favored by 25 Pentagon “power users.” That points to hands-on operational testing rather than a procurement announcement. The reports do not describe a final winner or a department-wide rollout. A separate federal track is also underway for classified model testing. (cnbc.com) Reports this week said Google, Microsoft and xAI agreed to let Commerce Department evaluators examine frontier models in classified conditions before public release. That is a different program from Pentagon operational use, but it shows federal agencies building channels to test commercial systems inside secure environments. (bloomberg.com) ### Where does xAI fit if OpenAI and Google are being tested? xAI already has a Pentagon foothold. Reuters, in a report carried by U.S. News, said the Pentagon has a $200 million deal with xAI and that Defense Secretary Pete Hegseth earlier this year announced the addition of Grok to GenAI.mil, the military’s generative AI platform. The same Reuters report said federal records outside the Pentagon showed far more examples of OpenAI-based tools than Grok in wider government use. (msn.com) CNBC reported in July 2025 that the Defense Department had awarded contracts of up to $200 million each to Anthropic, Google, OpenAI and xAI for AI development. That means the current testing is happening inside an existing structure of multiple Pentagon relationships, even as Anthropic’s standing changed later. ### Why are senators tying this to the drone budget? (usnews.com) The Senate Armed Services subcommittee raised the issue on May 20 as Pentagon officials defended the autonomous-warfare push. Military Times reported that Sen. Joni Ernst said the department’s “policy architecture really has to scale” with the budget request. Defense One said DAWG absorbed the earlier Replicator effort and now sits at the center of the Pentagon’s autonomy plans. (cnbc.com) Breaking Defense reported that only $1 billion of the $54.6 billion request sits in the regular base budget, with the rest tied to a more flexible reconciliation pot. Congress will decide whether to approve that request in the fiscal 2027 process, while the Pentagon continues model testing and vendor selection across OpenAI, Google, xAI and other contractors. (breakingdefense.com) (militarytimes.com)