Are we in check with what's out there? — Standards Gap-Check
Internal · 2026-06-07 · grounded in live mid-2026 sources (not memory). Answers Richard's question: "You mentioned there were protocols out there — are we missing anything, are we in check?"
Short answer
Two different scorecards, two different verdicts.
- On IT / cybersecurity governance — yes, we're in check (with two version bumps to make). The MCG rationalization engagement already anchors to the right, current frameworks: NIST CSF 2.0 and CIS Controls (update the reference to v8.1). Add PCI DSS v4.0.1 and the ~19–20 active US state privacy laws and we're current.
- On AI governance — no, there's a whole surface missing. The existing assessment treats the AI model as a black-box asset. It has zero AI-specific governance. That's not a small gap — it's the entire reason this product exists. Below is exactly what's missing and which recognized standard each piece comes from, so we're operationalizing what's out there, not inventing.
What the existing assessment already covers (and is current on)
| Area | Standard it uses | Status (June 2026) | Action |
|---|---|---|---|
| Cybersecurity posture | NIST CSF 2.0 (Govern function added vs 1.1) | Current | ✅ keep |
| Security controls | CIS Controls v8 | v8.1 is current (153 safeguards + CSF 2.0 mapping) | 🔁 bump to v8.1 |
| Payment data | PCI DSS | v4.0.1 (v4.0 retired end-2024; new reqs mandatory since Mar 2025) | 🔁 cite v4.0.1 |
| Privacy | State laws, GDPR, PIPEDA, FERPA, HIPAA | ~19–20 US states active 2026; HIPAA Security Rule update in NPRM, not final | 🔁 refresh the state list; flag HIPAA NPRM |
| Attestation | SOC 2 (Trust Services Criteria) | Current | ✅ keep |
Verdict: in check on security. Minor freshness edits, no structural gap.
What's missing — the entire AI-governance surface
NIST CSF 2.0 + CIS + SOC 2 govern infrastructure, identity, data, and operations. They do not cover model behavior, model lifecycle, or AI-specific failure modes. An org can pass every security control and be completely uncovered here. These are the eleven domains a security-only assessment structurally misses — and the recognized standard each lives in:
| # | AI-governance domain (missing) | Lives in / standard ref |
|---|---|---|
| 1 | Model bias / fairness testing | NIST AI RMF (Measure) · ISO/IEC 42005 impact assessment · EU AI Act (high-risk) |
| 2 | Prompt-injection / jailbreak defense | OWASP LLM01 · MITRE ATLAS · UK AISI red-team |
| 3 | Model inventory / AI-BOM (models, datasets, prompts, fine-tunes) | NIST AI RMF (Map) · ISO/IEC 42001 |
| 4 | Data provenance / training-data lineage | ISO/IEC 42001 · NIST AI 600-1 · EU AI Act data-governance |
| 5 | Human oversight / HITL | EU AI Act Art. 14 · NIST AI RMF (Govern) · OWASP LLM06 |
| 6 | Model drift / performance monitoring | NIST AI RMF (Manage) · ISO/IEC 5338 lifecycle |
| 7 | Agentic-system controls (tool-use, excessive autonomy/permissions) | OWASP LLM06 Excessive Agency · MITRE ATLAS · CSA Agentic profile |
| 8 | Third-party / foundation-model risk | EU AI Act GPAI obligations · ISO/IEC 42001 supplier controls · Databricks DASF |
| 9 | AI acceptable-use policy | ISO/IEC 42001 · NIST AI RMF (Govern) |
| 10 | Explainability / transparency | EU AI Act transparency (full Aug 2026) · NIST AI RMF · CHAI model cards |
| 11 | AI-specific incident response (hallucination harm, jailbreak incidents, EU serious-incident reporting) | EU AI Act incident reporting · NIST AI 600-1 · MITRE ATLAS |
The minimum credible AI-governance overlay the platform should map every control to:
- NIST AI RMF 1.0 + Generative AI Profile (NIST AI 600-1) — US voluntary baseline (the four functions: Govern/Map/Measure/Manage).
- ISO/IEC 42001:2023 — the certifiable AI Management System ("ISO 27001 for AI") — the anchor for any future govrn certification.
- OWASP LLM Top 10 (2025) + MITRE ATLAS (v5.1.0) — the technical AI threat surface.
- EU AI Act — wherever there's any EU nexus (full applicability Aug 2, 2026 — months away).
- Sector overlays — for health: The Joint Commission + CHAI model cards (see correction below).
Corrections to my own earlier vet (intellectual honesty)
The standards moved since my knowledge cutoff; the live research caught three things I'd have gotten wrong:
- 🔴 CHAI's "AI Assurance Labs" were SCRAPPED. In my partnership vet I cited CHAI as occupying the health-AI standard/assurance tier. That national assurance-lab network never materialized and was abandoned. What survives: CHAI's Applied Model Card ("nutrition label"), a public registry, and Governance Playbooks (May 2026). Real regulatory momentum shifted to The Joint Commission (accredits ~80% of US hospitals; partnered with CHAI June 2025, issued responsible-AI guidance Sept 2025). → For health-AI positioning, lead with Joint Commission + CHAI model cards, not assurance labs.
- US AI Safety Institute → CAISI. The US AISI was rebranded to NIST's Center for AI Standards and Innovation (CAISI) with a smaller mandate. Cite CAISI, not AISI, for US.
- ISO/IEC 27090 (AI security) & 27091 (AI privacy) are NOT final yet (FDIS / DIS as of early-mid 2026). They'll close the AI-security/AI-privacy bridge when published — do not claim conformance to them today. Same caution on OWASP ML Security Top 10 (still draft v0.3).
So: are we in check?
- Security: yes — current frameworks, two version bumps (CIS v8.1, PCI v4.0.1) and a privacy refresh.
- AI governance: no — and that's the product. We are not missing anything relative to the established AI standards once we build the overlay above; the existing assessment is missing the whole overlay. The platform's job is to operationalize NIST AI RMF + ISO 42001 + OWASP/ATLAS + EU AI Act into living, monitored controls — which is exactly what
FRAMEWORK.mdlays out.
Every standard/version/status above is from live sources pulled 2026-06-07; full source list in the research appendix. Items tagged uncertain there (NIST AI 600-1 April-2026 revision date; exact ATLAS counts) are flagged — verify before putting in a client-facing claim.