After Washington bars Anthropic, the Pentagon pivots vendors, exposing AI lock-in risks and accelerating interest in open and auditable models for mission continuity.
Federal restrictions on proprietary AI are reshaping US defence technology strategy, with procurement teams increasingly weighing open and source-available alternatives to avoid operational disruption.
The Trump administration has barred Anthropic from federal use, prompting the Pentagon to reportedly shift deployments to OpenAI. The move carries immediate procurement consequences and highlights a deeper concern: foundation models are becoming infrastructure, not interchangeable software.
According to Ben Van Roo, CEO of Legion Intelligence, demand will not slow.
“Once embedded in operational systems, foundation models behave less like software subscriptions and more like infrastructure dependencies. If this standoff continues, procurement will not stall. It will reallocate within the quarter. Demand inside the national security system does not disappear because a policy team is uncomfortable.”
He added that current military use is analytic rather than autonomous: “In practice, these models are not being used for autonomous lethal action. They are used to accelerate human analysis, surface intelligence gaps, and reduce decision latency. The public debate assumes autonomous weapons. The operational reality is analytic acceleration.”
With active US and allied strikes on Iranian targets compressing timelines, capability gaps carry immediate consequences.
“In live crises, capability gaps are not theoretical. They change outcomes.”
The episode also raises governance concerns, as product-level restrictions by private labs effectively define lawful government use. Meanwhile, adversarial model distillation and open-source replication weaken unilateral controls, strengthening the case for auditable, self-hosted, interoperable AI stacks.
For defence buyers, continuity now outweighs brand loyalty.
