Compliance FrameworksAmericas
NIST SP 800-218A
SSDF AI Supplement: AI-specific software development practices and supply chain controls.
SSDF AI Supplement: AI-specific software development practices and supply chain controls.
| Short name | 218A |
| Region | Americas |
| Country or scope | United States |
| Sector | AI |
| In your workspace | On unless you switch it off in Compliance โบ Manage. |
How Safeguard covers 218A
- Scan requirements: 6, all decided by a scan.
- Controls: the scan requirements below, plus 11 controls from Safeguard's control catalog for 218A.
- Automated tests: none map to it directly. You can link any automated test to its controls yourself (how).
- Policy templates: 25: 3 written for 218A and 22 from the standard set.
How each kind of coverage works, and what the readiness score counts: Compliance frameworks.
What a scan checks
Each scan of a product checks these 6 requirements against what the scan found, including the product's software inventory (SBOM).
| Requirement | What Safeguard checks | Decided by |
|---|---|---|
SP800218A-PS.3.2 Provenance Data / SBOM | Provenance data for all components of each software release must be collected, safeguarded and shared, e.g. as an SBOM in a standard machine-readable format (SP 800-218A profile of SSDF PS.3.2) | Scan |
SP800218A-PS.2.1 Integrity Verification Information | Integrity verification information such as cryptographic hashes must be made available to software acquirers (SP 800-218A profile of SSDF PS.2.1) | Scan |
SP800218A-PW.4.1 Well-Secured Component Reuse | Reused components, including AI frameworks and model artifacts, must be acquired from evaluated sources, presupposing a named supplier for each component (SP 800-218A profile of SSDF PW.4.1) | Scan |
SP800218A-PW.4.4 Component Vulnerability Verification | Acquired components must be verified for known vulnerabilities before use, which requires knowing each component's exact version (SP 800-218A profile of SSDF PW.4, component verification) | Scan |
SP800218A-RV.1.1 Vulnerability Monitoring | Vulnerability databases and sources must be monitored for issues affecting the software and its components, requiring identifiable components via PURL or CPE (SP 800-218A profile of SSDF RV.1.1) | Scan |
SP800218A-PS.3.2-PROV Release Provenance Recording | Provenance collection includes recording who or what produced the composition data for each release (SP 800-218A profile of SSDF PS.3.2, provenance data) | Scan |
More controls in the catalog
The framework's Controls tab lists its scan requirements and these controls from Safeguard's control catalog. Evidence them with automated tests, uploaded evidence, policies or an attestation.
| Control | Name |
|---|---|
SSDF-AI-PO.1 | Define security and safety requirements for AI-enabled software |
SSDF-AI-PO.2 | Implement secure and trustworthy AI development environment |
SSDF-AI-PO.5 | Assess and manage AI supply chain risks |
SSDF-AI-PS.1 | Design AI software with security and safety in mind |
SSDF-AI-PS.2 | Verify and validate AI model integrity and outputs |
SSDF-AI-PS.3 | Secure training data and protect against data poisoning |
SSDF-AI-PW.1 | Use secure practices for AI model development and training |
SSDF-AI-PW.4 | Reuse vetted AI components and pretrained models responsibly |
SSDF-AI-RV.1 | Monitor and respond to AI-specific vulnerabilities in production |
SSDF-AI-RV.2 | Analyze root causes of AI system failures and vulnerabilities |
SSDF-AI-RV.3 | Remediate AI vulnerabilities and communicate to affected parties |
Policy templates
Written for 218A:
- AI/ML Model Supply Chain Security Policy
- Secure Development Practices for AI-Enabled Software Policy
- AI/ML Vulnerability and Incident Response Policy
And 22 from the standard set, adapted to 218A.