# What "US-Only" Means for AI Model Providers: Four Layers to Check October 30, 2026 ยท Eric Magliarditi Canonical: https://consus.io/blog/what-us-only-really-means-for-ai-model-providers > US-only can mean who built the model, where it runs, who touches your data, or what happens to it after. How to check all four, and how Consus covers them. When government contractors ask for US-only AI model providers, they usually mean one of four different things. Mixing them up is how a team ends up with a vendor that sounds right on a slide and fails on a diligence call. "US-only" can describe who built the model, where the model runs, who can touch your data, and what happens to your data afterward. A serious evaluation checks all four. (For how US-only fits into a CMMC assessment more broadly, see [US-Only AI Models and CMMC Compliance](/blog/us-only-ai-models-and-cmmc-compliance-for-defense-contractors).) This post walks through each layer and shows how Consus gives defense contractors a US-only route to leading AI models without making them assemble the pieces themselves. ## Layer 1: Who is the model provider? The first meaning of US-only is the simplest: the company behind the model. Consus routes requests to models from OpenAI, Anthropic, Google, and SpaceXAI, so you get a choice of leading US model makers through a single API instead of negotiating a separate relationship with each one. The current list, with the authorization level each model supports, is in the [Consus model catalog](https://gateway.consus.io/models). ## Layer 2: Where does the model run? A US-based provider can still serve your request from the wrong place. For "US-only" to mean anything to a government contractor, the infrastructure has to be as controlled as the company name. Every Consus request stays inside a US-only, FIPS-encrypted boundary in US-only cloud regions: AWS GovCloud, GCP Assured Workloads, and Azure Government. There is no commercial-cloud routing path in the system, and no silent fallback if a model is unavailable at the compliance level you asked for. The request fails instead of drifting somewhere you did not approve. ## Layer 3: Who can touch the data? For export-controlled work, location is not enough. You also need to know about the people. Everyone with production access to the Consus data plane is a US person as defined in 22 CFR 120.62, and that access is logged and reviewed. Requests are encrypted in transit with FIPS 140-2 validated endpoints and are processed only in US cloud regions. Be precise about what that does and does not mean. You remain the registered exporter for your own technical data, and the export determinations stay with you. Consus provides the controlled path. The decisions that are yours stay yours. ## Layer 4: What happens to the data afterward? This is the layer most often skipped. A US provider on a US cloud still creates risk if it keeps your prompts. With Consus, prompts and completions never persist. Nothing you send is stored by Consus, and zero data retention is in place with every upstream model provider. Content lives only in memory while a request is in flight. The only data kept is operational metadata, such as the model, token counts, and timestamps. ## Why a gateway beats stitching it together You could build all four layers yourself: sign separate agreements with each model provider, stand up government-cloud accounts, federate identity, configure encryption and logging, and document every data flow for your assessor. It can be done. It is also a long project, and parts of it repeat every time you add a model. A US-only gateway collapses that work. With Consus, the compliance level is part of every request, through a model ID suffix such as `:fedramp-high`, `:il5`, or `:itar`. A request with no suffix is rejected. Consus is OpenAI compatible, so most tools only need a new base URL, API key, and model ID. It works with the OpenAI SDK, LangChain, LlamaIndex, LiteLLM, Claude Code, Codex CLI, Cline, OpenCode, and VS Code Copilot. ## What US-only does not do US-only is a boundary, not a certification, and we say so. Consus does not classify your content or detect CUI markings. It does not stop a user from choosing the wrong compliance level for their data. It does not secure your client applications or API keys. And no single vendor makes a contractor CMMC compliant, because CMMC assesses your whole system. What Consus does provide is evidence. Consus is FedRAMP Equivalent, assessed at High: our 3PAO tested Consus against all 410 FedRAMP High controls, and the Body of Evidence for your CMMC assessor is available under NDA. ## A quick diligence checklist When you evaluate US-only AI model providers for government work, ask these questions of every option: - **Provider:** Which companies' models are available, and how current is the list? - **Hosting:** Which cloud regions serve requests, and is there any fallback outside them? - **People:** Who operates the data plane, and is access logged and reviewed? - **Retention:** Is any content stored by the gateway or by the model provider? - **Boundary:** What is the vendor responsible for, and what is left to you? - **Evidence:** What can your assessor review, and under what terms? Consus has a documented answer for each one. ## The bottom line US-only is not one property. It is four, and they all have to hold at once. Consus gives government contractors a US-only route to models from OpenAI, Anthropic, Google, and SpaceXAI, running in controlled US cloud regions, handled by US persons in the data plane, and never stored along the way. Ready to check all four layers? Visit [consus.io](https://consus.io) to request access or book a compliance call. Access is reserved for US-based defense contractors and follows a company review and contract.