Ten Questions an Assessor Will Ask About Your Generative AI Tools
If your team uses generative AI tools, expect your CMMC assessor to ask about them. Using generative AI tools without violating CMMC requirements is not about having the right opinions on AI. It is about answering plain questions with documentation instead of assurances.
This is not an official assessor checklist, and every assessment is different. But these ten questions cover the ground a careful assessor is likely to walk. For each one, here is how Consus gives defense contractors a clear, documented answer.
1. Where does our data go when someone uses a generative AI tool?
With Consus, every request stays inside a US-only, FIPS-encrypted boundary and is routed to AWS GovCloud, GCP Assured Workloads, or Azure Government. There is no commercial-cloud routing path in the system, so there is no stray route to find.
2. How do you stop CUI from reaching the wrong environment?
Consus puts the compliance level in the model ID itself. Every request carries a suffix such as fedramp-high, il5, or itar. A request with no suffix is rejected with a 400 error before it reaches any model. There is no default compliance level and no silent fallback to one.
3. Is any of our content stored?
No. Prompts and completions never persist. Nothing sent through Consus is stored by Consus, and every model provider on every route is configured for zero data retention. Inside Consus, content lives only in ephemeral container memory while a request is in flight. The only data kept is operational metadata such as the model, token counts, and timestamps.
4. How is data protected in transit?
TLS 1.2 or higher on every connection, with FIPS 140-2 validated endpoints by default.
5. How do the services authenticate to each other?
Consus uses Workload Identity Federation to reach GCP and Azure, exchanging short-lived tokens instead of storing long-lived cloud credentials. There are no static GCP or Azure keys sitting at rest for someone to find.
6. How is the network isolated?
The gateway runs in private subnets with no direct internet access. Customer requests enter only through a single controlled front door, and traffic to AWS services stays on private links.
7. What do your logs contain, and how long are they kept?
Logs contain metadata only, never prompt or response content, and they are KMS-encrypted. AWS API Gateway access logs and application logs are kept for 365 days, Azure diagnostic logs for 365 days, and GCP audit logs carry a three-year retention lock.
8. What about AI agents and coding assistants?
Agentic generative AI tools can call tools that reach outside the conversation, so Consus applies guardrails:
- Hosted tools are blocked. Server-executed tools such as web search and hosted code execution never reach the model.
- Exfiltration-shaped tools are rejected. Tool schemas that describe an outbound destination, such as a webhook or callback URL, are rejected.
- Suspicious tool calls are flagged. Tool calls that contain an outbound destination are flagged in the response and logged.
Consus never executes tool calls. Your tools run in your environment.
9. What is the vendor responsible for, and what are we responsible for?
Consus draws this line openly. It does not classify content, it does not detect CUI markings, it does not stop a user from choosing the wrong compliance level for their data, and it does not secure your client applications or API keys. Those stay with your team. A clean, documented boundary is exactly what an assessor wants to see.
10. Where is the evidence?
Consus is FedRAMP Equivalent, assessed at High: a third-party assessment organization (3PAO) tested Consus against all 410 FedRAMP High controls. You can fold that documentation into your own system description instead of writing the gateway layer from scratch. The Body of Evidence package is available under NDA.
Making the answers easy
Notice what these answers have in common. Most of them are not policies your people have to remember. They are properties of the system. The compliance level is in the request. The commercial route does not exist. The content is not stored. That is the difference between a team that hopes it is using generative AI tools without violating CMMC requirements and a team that can show it.
For the full walkthrough of what CMMC requires of an AI workflow, see How to Use Generative AI Without Violating CMMC.
Consus is also OpenAI compatible, so adopting it usually means changing a base URL, an API key, and the model ID. It works with the OpenAI SDK, LangChain, LlamaIndex, LiteLLM, Claude Code, Codex CLI, Cline, OpenCode, and VS Code Copilot.
The bottom line
CMMC requirements do not ask you to give up generative AI tools. They ask you to know where your data goes and to prove it. Consus gives your team the frontier models it wants, through the tools it already uses, with a documented answer ready for every question above.
Want to see the evidence? Visit 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.