AI Security Framework Crosswalk

How the major frameworks fit together: govern, model the threats, select controls, harden & operate

These frameworks aren't competitors; they're layers. Governance standards sit on top, threat models tell you what to defend, control catalogs say what to implement, and hardening baselines make it real. Most pair explicitly with one another.

How they layer

① Govern

Org-level accountability & risk management

ISO/IEC 42001 (AI management system) NIST AI RMF: Govern function

▾

② Model the threats

Enumerate what can go wrong

OWASP LLM Top 10 + Agentic (ASI) MITRE ATLAS (adversary techniques) CSA MAESTRO (7-layer method)

▾

③ Select controls

Choose & specify auditable safeguards

CSA AI Controls Matrix (243 controls; 247 in v1.1) NIST 800-53 / CSF 2.0 ISO 42001 Annex controls

▾

④ Harden & operate

Implement the basics well; enforce Zero Trust

CIS Controls v8.1 + Benchmarks NIST SP 800-207 (Zero Trust)

Reference matrix

Framework

Maintainer

What it gives you

Connects with

NIST AI RMF+ AI 100-2, CSF 2.0, 800-207

NIST

GovernGovern / Map / Measure / Manage risk functions, an adversarial-ML taxonomy (AI 100-2), and the U.S. federal control spine.

Governance peer to ISO 42001; AICM maps to it; CIS maps to its CSF.

ISO/IEC 42001ISO/IEC · certifiable

ISO/IEC

GovernThe AI management system standard: the "ISO 27001 for AI." A certifiable, organization-wide governance backbone.

Pairs with NIST AI RMF; AICM maps to it. (Anthropic reports it was among the first AI firms certified.)

OWASP GenAILLM Top 10 + ASI

OWASP

ThreatsThe shared threat vocabulary: LLM Top 10 (app-level) + Agentic Security Initiative (ASI Top 10). Also AI-BOM, and coined "least agency."

Pairs with ATLAS techniques; its multi-agent guide is built on MAESTRO.

CSA MAESTROCloud Security Alliance

CSA

ThreatsControls7-layer threat-modeling method for agents, plus the AI Controls Matrix (243 controls / 18 domains; 247 in v1.1) and a STAR-for-AI path.

AICM maps to ISO 42001 + NIST AI RMF + BSI AIC4; feeds the OWASP multi-agent guide.

CIS Controls+ CIS Benchmarks

CIS

OperateThe prioritized hardening baseline: 18 Critical Security Controls (v8.1) and CIS Benchmarks for config hardening, the operational floor under every AI control.

Maps to NIST CSF; underpins the Zero Trust controls in the other layers.

Key mappings to remember

CSA AI Controls Matrix → ISO 42001 + NIST AI RMF + BSI AIC4

OWASP multi-agent threat-modeling guide ← built on CSA MAESTRO

CIS Controls → NIST CSF 2.0 (and most major frameworks)

NIST SP 800-207 Zero Trust → underpins Anthropic's Zero Trust for AI Agents