OpenAI Moves Into Classified Pentagon Work as Anthropic Is Frozen Out

OpenAI is moving deeper into the U.S. national security apparatus, extending its Pentagon work into classified environments and tapping Amazon’s powerful government cloud as rival Anthropic fights for survival on the outside. In late February, OpenAI announced a deal with the Department of Defense to deploy its models on the military’s classified network, following Anthropic’s refusal to support autonomous weapons and mass surveillance uses. OpenAI’s blog framed the arrangement as adding “more guardrails than any previous agreement,” while reporting from Al Jazeera and others underscored that the systems will sit inside a secure, classified environment.

The move lands just months after OpenAI inked a seven‑year, $38 billion cloud pact with Amazon Web Services, giving it massive access to AWS infrastructure that already underpins major U.S. intelligence and defense workloads. The Guardian reported that the AWS agreement lets OpenAI run its frontier models on Amazon data centers worldwide, backed by “hundreds of thousands” of Nvidia GPUs, while Amazon’s own summary touts joint work to bring OpenAI models to AWS customers at scale.The Guardian Amazon With AWS already hosting classified government clouds, that combination positions OpenAI to sell both unclassified and classified services into the federal market.

Anthropic, by contrast, has been pushed to the margins of Pentagon work after initially being the lab whose Claude model was embedded in classified missions via Palantir and AWS. In early March the Defense Department formally labeled Anthropic a “supply‑chain risk,” a designation usually reserved for foreign adversaries, forcing defense contractors to rip out Claude from projects tied to the department, as detailed by TechCrunch and legal analysis from Mayer Brown. Anthropic has responded with a federal lawsuit arguing the label is unconstitutional and retaliation for refusing “unrestricted” military use, according to Axios.

The result is a dramatic reshaping of who gets to build core AI for the U.S. defense state. Scale AI, Microsoft and others already run models on classified networks, but OpenAI’s pivot from a public stance against military use to being a premier classified vendor—while a safety‑focused rival is effectively blacklisted—signals consolidation of power among firms willing to align with Pentagon definitions of “lawful” AI use.WIRED