FedRAMP Authorization Is Not Your Homework Done: Shared Responsibility for SMB AI and Cloud
FedRAMP authorization and in-scope AI services like Amazon Bedrock do not finish the customer's security work. Shared responsibility lessons for SMB cloud and AI buyers.
Sales decks love a badge. “FedRAMP Authorized.” “In scope.” “Government-ready AI.” Those labels matter — and they are easy to misunderstand. For small and mid-size businesses buying cloud or generative AI, the badge answers what the provider secured in the platform. It does not answer whether your tenant, prompts, identities, and data paths are under control.
AWS states the point plainly in its Shared Responsibility Model: security and compliance are shared. AWS protects security of the cloud (infrastructure). Customers remain responsible for security in the cloud — including data, identity and access management (IAM), guest configuration where applicable, and how services are integrated into the business.
As AI services show up on FedRAMP services-in-scope lists — Amazon Bedrock model families, for example, with Class C (formerly Moderate) in U.S. East/West and Class D (formerly High) in GovCloud on AWS’s model status page — the confusion scales. A regulated or enterprise customer may require an authorized AI path. Your team still owns prompts, retrieval data, keys, logging, and who can call the API.
What Shared Responsibility Means in Practice
- Inherited: Physical and environmental controls at the provider facility — you do not rebuild the data center.
- Shared: Patching and configuration — provider patches infrastructure; you patch what you run and configure what you own.
- Customer-specific: Your applications, data classification, encryption choices, IAM policies, and network zoning decisions.
The more abstracted the service (object storage, managed AI APIs), the less OS work you do — and the more your remaining work concentrates on identity, data, and application design. That is not “less security.” It is a different checklist.
The SMB Buyer Checklist (Non-Federal)
- Badge vs. tenant: Ask vendors which authorization covers the exact region and service you will use — then ask what you must still configure on day one.
- Identity: Who can invoke the AI/cloud admin APIs? Are standing admin roles minimized? Is MFA enforced on those accounts?
- Data path: Where do prompts, embeddings, and customer files live? Who can export them? What retention did you choose?
- Logging: Are API and admin logs enabled and retained somewhere a human reviews? Authorization does not read your logs for you.
- Contracts: If an MSP or integrator builds the AI stack, require a written shared-responsibility matrix — provider / integrator / your company.
This week’s action: Pick one cloud or AI service you rely on. Write a half-page: what the provider secures, what your team (or MSP) secures, and the three settings that would hurt most if misconfigured. Date it. That is your shared-responsibility starter artifact.
How This Differs from Asking for a Secure Configuration Guide
Secure configuration guides explain knobs and defaults inside a product. Shared responsibility explains whose job the knobs are. You need both conversations. A FedRAMP badge without an ownership map is how “we thought the cloud was secure” becomes an incident postmortem.
Soft Next Step
CyberIntelPro / Security Medic drafts a right-sized shared-responsibility matrix for Microsoft 365, AWS, or an AI pilot, as a one-pager built with your MSP or internal team.
References
- AWS, Shared Responsibility Model: aws.amazon.com/compliance/shared-responsibility-model/
- AWS, Amazon Bedrock models — FedRAMP and DoD CSP SRG certification status: aws.amazon.com/…/amazon-bedrock-models/
- AWS, FedRAMP Services in Scope: aws.amazon.com/compliance/services-in-scope/FedRAMP/