The Human Assisted Intelligence Assurance Maturity Model (HAIAMM) provides a comprehensive framework for organizations designing and implementing AI to automate workflows and augment capabilities, what we call Human Assisted Intelligence (HAI).
HAIAMM addresses the governance, building, verification, and operations of HAI systems with foundational practices to ensure trust, safety, and security.
Traditional security frameworks (ISO 27001, NIST CSF) weren't designed for Human Assisted Intelligence deployments. HAIAMM fills this gap by providing:
| Version | Date | Changes |
|---|---|---|
| v3.0 CURRENT | 2026-05-14 |
Subject reframing, AI/HAI as subject, not tool, The subject of every (domain × practice × level) cell is now the AI/HAI systems, data, infrastructure, workflows, endpoints, and vendor tools the organization builds, consumes, or operates. AI-as-security-tool framing (AI-SOC, AI-DLP, AI vendor scoring) retired 216 canonical cells, 6 domains × 12 practices × 3 levels, each authored to the canonical template (Practice Overview → per-level Objective/Dependencies/Outcomes/Activities/Outcome Metrics/Process Metrics/Effectiveness Metrics/Success Criteria → trailer with Pitfalls and Maturity Questions) 72 one-pagers at full conformance, Vendors, Software, Data, Infrastructure, Processes, Endpoints, all 12 practices each HAI-specific threat taxonomy, EA (Excessive Agency), AGH (Agent Goal Hijack), TM (Tool Misuse), RA (Rogue Agents) elevated from footnote to first-class category, threaded through TA → SA → ST → ML Vendors as a first-class domain with shadow AI reduction as the primary L1 outcome, full 12-practice, 3-level treatment for vendor-provided AI and AI-embedded SaaS MITRE ATLAS canonical, elevated from "one reference" to the canonical adversarial-ML reference; ATLAS tactics TA0001-TA0014 walked per archetype in TA; AML.M00xx mitigations referenced in SA; AWS per-cloud threat-model template authored (Azure/GCP pending) Outcome metrics by default, every level prescribes outcome metrics with baseline, target, and source. No activity without a metric Risk-tier-driven calibration, SM L2 produces a risk-tier rubric and tier-treatment matrix in each domain; every other L2 cell inherits Priority compliance map, EU AI Act (Art. 9/12/14/15/26/50, Annex III), NIST AI RMF + Playbook (GOVERN/MAP/MEASURE/MANAGE), GDPR (Art. 22/28/32/33, 44-49), ISO/IEC 42001, ISO/IEC 27001 (A.5/A.8), SOC 2 CC9.2, plus HIPAA/PCI-DSS/FINRA/SEC/HHS-FDA/NYDFS Part 500/OCC/NYC LL 144/CO SB-21-169/FCRA/EEOC/FERPA/COPPA 72 questionnaires rewritten to v3.0 with outcome-metrics scoring v3.0 Handbook (PDF), full practitioner reference available for download |
| v2.0 | 2025 |
Assessment methodology, Comprehensive questionnaire-based assessment (OpenSAMM v1.0 methodology), scoring, 5-phase process, industry benchmarks Threat Intelligence, Elevated to foundational capability across all 6 domains (consumption → analysis → production) Prompt Injection Security (Arcanum PI Taxonomy by Jason Haddix, CC BY 4.0) integrated across 15 practice one-pagers: 13 attack intents, 18 techniques, 20 evasion methods Critical HAI Assurance: 4 agentic AI risks (EA/AGH/TM/RA), “Least Agency Principle” governance, 21 new practice-domain combinations, 47 new assessment questions, 95% OWASP alignment (LLM Top 10 2025 & Agentic Top 10 2026) Note: v2.0 framed AI as a security tool (AI doing SAST/DLP/SOC/vendor scoring). This subject framing is retired in v3.0. |
| v1.0 | 2024 | Initial framework release, 12 practices, 6 domains, 3 maturity levels, 72 practice-domain combinations |
HAIAMM is open source. Visit the GitHub repository to contribute or submit issues. The complete HAIAMM Handbook v3.0 (PDF), Executive Summary, and Model Master Document are available on the downloads page.
Information security management alignment.
Cybersecurity Framework risk management.
AI Risk Management Framework governance.
Software assurance maturity model methodology.
95% coverage of LLM application risks.
95% coverage of agentic AI risks.