OpenAI GPT, GPT OSS and NVIDIA Nemotron models earn FedRAMP High and DoD IL-4/5 in Bedrock GovCloud

AWS expanded the compliance footprint of Amazon Bedrock, announcing that OpenAI GPT, OpenAI GPT OSS and NVIDIA Nemotron models are now FedRAMP High and DoD CC SRG IL-4/5 approved inside AWS GovCloud (US). That authorization is the gate that lets federal agencies, defense organizations and other public-sector buyers with strict compliance mandates run these frontier and open models in production rather than just pilots. In a parallel move, AWS said its agentic engineering partner tool Kiro also achieved FedRAMP High and DoD IL-4/5 authorization in GovCloud.

The mechanics matter: FedRAMP High and IL-4/5 are among the most demanding US government cloud accreditations, covering sensitive and controlled-unclassified workloads. Clearing them inside Bedrock means agencies can use familiar AWS procurement and security controls instead of bespoke deployments — a major reduction in the friction that traps government AI in experimentation.

This lands squarely inside the week's policy theme. As Washington tightens who can access frontier models (the GPT-5.6 stagger, Anthropic's export restrictions), AWS is simultaneously making approved models broadly available to the government itself — federal access is expanding even as commercial and foreign access narrows. It also strengthens AWS's enterprise/gov pitch against Azure (which leans on its own OpenAI relationship) and Google's gov cloud.

Competitive context: Bedrock adoption has surged, with AWS citing 170% sequential spend growth in Q1 2026 and use by ~80% of the Fortune 100. The gov accreditation deepens that moat. Caveat: authorization is necessary but not sufficient — actual defense/agency adoption, data-handling rules and model-update recertification will determine impact. Watch which agencies deploy first and whether Anthropic's restricted models can ever clear the same bar.