NIST SP 800-53 control overlays for securing AI systems (COSAiS): the five use cases and what is still draft

Why this page: defense contractors who already map controls to NIST SP 800-171 ask what the equivalent is for the AI system itself. NIST’s answer in progress is a set of SP 800-53 overlays. This page reports what NIST’s project pages say as fetched on 16 October 2026. The project is a work in progress, so nothing here is a requirement.

What an overlay is

NIST’s FAQ says Control overlays enable organizations and communities of interest to customize the controls (or control baselines) for a specific technology, system, mission space, and environment of operation. Controls can be selected from the SP 800-53 control catalog, modified to address unique risks or applications, and supplemented to provide application-specific guidance for implementers.

On whether AI needs different controls, the FAQ says The controls needed to manage risks to AI systems and components will be similar to those required for any type of software.

The five planned use cases

The project overview says the overlays will develop a series of overlays for securing AI systems using the NIST Special Publication (SP) 800-53 controls and lists these proposed use cases:

1. Adapting and Using Generative AI – Assistant/Large Language Model (LLM)
2. Using and Fine-Tuning Predictive AI
3. Using AI Agent Systems (AI Agents) – Single Agent
4. Using AI Agent Systems (AI Agents) – Multi-Agent
5. Security Controls for AI Developers

The overview says the overlays will also leverage NIST SP 800-218A, Draft NIST AI 800-1, and NIST AI 100-2e2025 and are focused on protecting the confidentiality, integrity, and availability of information and users. For the SSDF side, see SP 800-218A for generative AI in defense software.

Status

The overview’s recent-updates list includes August 14, 2025: The NIST SP 800-53 Control Overlays for Securing AI Systems Concept Paper is available for comment and, for January 2026, an annotated outline (discussion draft) of Control Overlays for Securing AI Systems: Using and Fine-Tuning Predictive AI. The publications page lists only Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations as a directly related NIST-authored publication at the time we fetched it, so we found no final overlay.

The FAQ says Organizations are not required to use the provided overlays. It calls them a starting point for cybersecurity considerations for organizations using AI and AI developers.

What to do while the overlays are draft (editorial judgment)

This section is the site’s judgment, not NIST’s.

• Keep the control baseline you are assessed on. The FAQ’s point that AI controls resemble software controls supports continuing with your existing baseline for CUI work. See CMMC Level 2 versus Level 3 for AI.
• Pick the use case that matches your tool. An LLM assistant, a fine-tuned predictive model and an agent have different component lists. Ask a vendor which of the five it considers itself.
• Do not cite a draft outline as a control set. Ask for the vendor’s own control mapping, and record which NIST document and date it used.

Questions

Are overlays part of CMMC?

Nothing on the NIST pages we fetched ties COSAiS to CMMC. The site’s CMMC pages cover the assessed requirements.

How do overlays relate to the AI RMF?

The FAQ says the overlays will be designed to address specific risks to different types of AI usage in conjunction with an organization’s cybersecurity risk management program and existing control implementation. The AI RMF is a separate NIST framework; see the AI RMF explainer.