What's New

OpenAI GPT, OpenAI GPT OSS, and NVIDIA Nemotron models on Amazon Bedrock have received FedRAMP High and DoD Cloud Computing Security Requirements Guide (CC SRG) Impact Level 4 and 5 approvals within AWS GovCloud (US) Regions. This makes these frontier and open-source models available to federal agencies, defense organizations, and public sector entities that must meet stringent government security and compliance mandates. The models are delivered through Mantle, Amazon Bedrock's next-generation distributed inference engine, ensuring high-performance, secure serverless inference.

How It Works

• Mantle Inference Engine: The models run on Mantle, a next-generation distributed inference engine built into Amazon Bedrock that provides high-performance serverless inference without requiring customers to manage underlying infrastructure.
• Zero Operator Access: Mantle enforces a zero-operator-access model, meaning no AWS personnel can access customer data or model inputs/outputs during inference, a critical requirement for government workloads.
• Automated Capacity Management: Mantle handles capacity scaling automatically, eliminating the need for manual provisioning and ensuring consistent performance under variable government workloads.
• OpenAI API Compatibility: The inference engine provides out-of-the-box compatibility with OpenAI API specifications, allowing agencies to migrate or integrate existing OpenAI-compatible tooling without code rewrites.
• AWS GovCloud (US) Isolation: All inference runs within the physically and logically isolated AWS GovCloud (US) Regions, which are restricted to US persons and US-based infrastructure to satisfy ITAR, FedRAMP, and DoD requirements.
• Compliance Authorization: The models have been formally authorized under FedRAMP High baseline and DoD CC SRG IL-4 (covering Controlled Unclassified Information) and IL-5 (covering National Security Systems data), enabling use with sensitive government data.
• Amazon Bedrock Security Controls: Standard Bedrock security features—including Guardrails, IAM-based access policies, and VPC integration—remain available to further harden government deployments.

Why It's Important

• Unlocks Frontier AI for Sensitive Government Data: Federal agencies can now use state-of-the-art OpenAI GPT and NVIDIA Nemotron models on data classified up to DoD IL-5, a threshold previously inaccessible with these commercial frontier models.
• Reduces Compliance Burden: Pre-authorized FedRAMP High and IL-4/5 status means agencies do not need to independently assess and authorize these models, significantly shortening the Authority to Operate (ATO) process.
• Enables Defense-Grade Generative AI Applications: DoD components, intelligence-adjacent agencies, and defense contractors can now build production-grade generative AI workflows—RAG pipelines, agents, summarization—on classified-adjacent data without leaving the GovCloud boundary.
• Broadens Model Choice in Regulated Environments: Previously, government customers were limited to a narrower set of compliant models; this expansion brings leading commercial and open-source models into the compliant portfolio.
• Supports Zero-Trust and Data Sovereignty Requirements: Zero operator access and GovCloud's US-person-only staffing model directly address zero-trust and data sovereignty mandates increasingly required by federal policy.
• Accelerates Public Sector AI Modernization: With compliance barriers removed, agencies can move faster from AI pilots to production deployments, supporting broader federal AI modernization initiatives.

How It's Different

• FedRAMP High + IL-5 Dual Authorization: Unlike standard commercial Bedrock regions, these models carry both FedRAMP High and DoD IL-5 authorization simultaneously, covering a wider range of government use cases in a single service.
• Mantle's Zero Operator Access vs. Standard Managed Services: Traditional managed AI services may allow provider personnel access for support or debugging; Mantle's architecture explicitly eliminates this, a differentiator for sensitive government workloads.
• OpenAI Models in a Sovereign Cloud: OpenAI GPT models are typically accessed via OpenAI's commercial API, which does not hold FedRAMP High or DoD IL-5 authorization; this announcement provides a compliant, sovereign alternative on AWS infrastructure.
• Serverless with No Capacity Planning: Unlike provisioned throughput options that require pre-commitment, Mantle delivers serverless inference with automated capacity management, reducing operational overhead for government IT teams.
• OpenAI API Spec Compatibility in GovCloud: The native OpenAI API compatibility within a FedRAMP High boundary is unique, allowing agencies to reuse existing OpenAI-compatible SDKs and integrations without modification.
• Bundled with Full Bedrock Ecosystem: Government users get access to Bedrock Guardrails, Knowledge Bases, Agents, and Data Automation alongside these models—a more complete platform than point-solution model APIs.

When to Prefer It

• Processing Controlled Unclassified Information (CUI): When an agency needs to run LLM inference on CUI data that requires IL-4 or IL-5 protection, these models are the appropriate choice over commercial-region alternatives.
• DoD and Intelligence Community Workloads: Defense organizations building AI-assisted decision support, document analysis, or mission planning tools that must remain within DoD CC SRG boundaries should use these models.
• Migrating from OpenAI Commercial API to a Compliant Environment: Organizations currently using OpenAI APIs in non-compliant environments who need to achieve FedRAMP High authorization can migrate to these Bedrock-hosted models with minimal code changes.
• Building FedRAMP High SaaS Products for Government: ISVs and system integrators building software products that will be sold to federal agencies and need to inherit FedRAMP High authorization from their cloud provider.
• Rapid ATO Scenarios: When a program office needs to deploy generative AI quickly and cannot afford a lengthy independent model assessment, leveraging pre-authorized models on Bedrock shortens the ATO timeline.
• Open-Source Model Deployment Without Infrastructure Management: When agencies want the flexibility of NVIDIA Nemotron open-source models but lack the infrastructure team to self-host them securely at IL-4/5, the serverless Bedrock option is preferable.
• Multi-Model Evaluation in a Compliant Context: When government data scientists need to benchmark OpenAI GPT, GPT OSS, and Nemotron models against each other using sensitive data, Bedrock's unified compliant environment enables side-by-side evaluation safely.

Availability

• Status: Generally Available (GA) as of June 25, 2026.
• Regions: Available in AWS GovCloud (US) Regions (us-gov-west-1 and us-gov-east-1); not available in standard commercial AWS regions under these compliance authorizations.
• Compliance Authorizations: FedRAMP High baseline and DoD CC SRG Impact Level 4 and Impact Level 5 approved.
• Models Covered: OpenAI GPT (frontier), OpenAI GPT OSS (open-source variants), and NVIDIA Nemotron model families on Amazon Bedrock.
• Inference Model: Serverless inference via the Mantle engine; provisioned throughput availability for these models in GovCloud should be confirmed via the Bedrock console or AWS account team.
• Pricing: Follows Amazon Bedrock's standard on-demand token-based pricing model; GovCloud pricing may differ from commercial regions and should be verified in the AWS GovCloud pricing documentation.
• Access Requirements: Requires an AWS GovCloud (US) account, which is restricted to US persons and US-based entities; standard AWS commercial accounts cannot access GovCloud resources.
• Limitations: Specific model versions and feature parity with commercial Bedrock regions (e.g., fine-tuning, batch inference) should be verified in the Amazon Bedrock GovCloud documentation, as not all capabilities may be available at launch.

Related Resources

• https://aws.amazon.com/bedrock/?trk=7ecf60df-6136-414c-a7c3-6aa4d2d6019f&sc_channel=ps&ef_id=EAIaIQobChMI0-nJtciglQMVh3JHAR3wnSV-EAAYASAAEgIIGvD_BwE&gads_camp=23532472972&gads_ag=194311072004&gads_ad=795877020842&gads_kw=amazon%20bedrock&gads_matchtype=e&gads_network=g&gads_device=c&gads_geo=9007742&gad_campaignid=23532472972&gclid=EAIaIQobChMI0-nJtciglQMVh3JHAR3wnSV-EAAYASAAEgIIGvD_BwE
• https://docs.aws.amazon.com/bedrock/
• https://docs.aws.amazon.com/govcloud-us/latest/UserGuide/govcloud-compliance.html