The Pentagon confirmed it is simultaneously running AI systems from Google, OpenAI, and other vendors without commercially available integration middleware cleared for DoD security requirements. GenAI.mil, the department's primary AI portal, has rapidly scaled to 1.3 million…
The Pentagon confirmed it is running AI from Google and OpenAI simultaneously — and has no integration software to manage the complexity that comes with juggling multiple vendors at once.
In an interview with CNBC published Monday, Cameron Stanley, the Defense Department's chief AI officer, confirmed that alongside Google, the department is working with OpenAI and other providers to modernize wartime capabilities. The interview landed alongside an uncomfortable data point: more than 700 Google employees sent a letter to CEO Sundar Pichai this week calling on the company to reject classified workloads, CNBC reported. Stanley had already put the vendor problem on record in the same interview. "Overreliance on one vendor is never a good thing," he said. "We're seeing that, especially in software."
The coordination challenge has a name inside the Pentagon: the middleware gap. The integration software that would let military users switch between AI providers, pool outputs, or manage which tool handles which task does not yet exist commercially in a form cleared for the department's security requirements, Gavin Kliger, the department's chief data officer, told Defense One this week. Brad Glassman, who leads AI adoption at the Defense Department, described the workforce dimension: the Deferred Resignation Program, the Trump administration effort that offered federal workers cash to leave, created a crisis. "We lost some really, really good people," he told DefenseScoop.
GenAI.mil, the Google-hosted portal housing the department's main AI operation, has accumulated 1.3 million active users out of a potential 3 million since launching December 9, reaching 1 million users within a month with no reported latency or downtime, Defense One reported. In five weeks, those users built more than 100,000 AI agents on top of it, DefenseScoop reported. One user at Navy Recruiting Command used Gemini to compress the time to build an automated personnel database from several years to three months, saving roughly 10 weeks of labor annually.
The build-or-buy question underneath is not a technicality. Whoever builds the integration layer that connects military users to multiple AI providers controls the workflow: a commercial platform vendor, a defense contractor, or the government itself. China has already chosen the government path, with state-directed coordination between labs and agencies and no procurement overhead between ambition and output. The Pentagon is assembling its version from commercial parts under political pressure, and nobody has yet signed up to own the thing that holds it together.
The number to watch in the next six months: whether any non-Google model actually appears on GenAI.mil, and whether it arrives before or after an employee lawsuit or contract renegotiation forces a structural change to the current arrangement.