Most companies govern AI in four separate places, and none of them can show the others did their job. Ethics sits in a principles document, law with counsel, risk in a framework spreadsheet, and controls inside engineering pipelines.
An integrated AI governance model connects the four so each principle traces to a legal duty, a NIST or ISO standard, and a control that leaves evidence. This paper reflects US and EU law as of November 2026, including Colorado's repeal and replacement of its AI Act and the EU's move of Annex III high-risk duties to December 2, 2027.
It is a synthesis of legal texts, standards and published surveys, not a study of any company's program, and it is not legal advice. Colorado rulemaking and federal preemption litigation are still moving.
Checking a governance program does not require reading every policy. It requires asking for three things that either exist or do not, and each one shows whether ethics, law, risk and engineering are working from the same list. Each question is answered by one named artifact.
Ask for the list of every AI system in use, including shadow AI and vendor features embedded in SaaS tools. Each entry needs an owner, purpose, model and version, data sources, users affected, jurisdictions, decision role and tier. Gartner calls a centralized inventory foundational to governance platforms, and without one no law can be mapped to a system. The artifact is the inventory, with a named owner on every row.
Ask which laws apply to each system, by jurisdiction and by date. Federal rules, each state where you operate or have users, NYC if you hire there and the EU if you sell there all belong on it. The dates are close: California CPPA ADMT compliance lands January 1, 2027, and EU Annex III high-risk duties apply from December 2, 2027. The artifact is an obligations register mapped to the control library.
several regimes
Ask to see the evidence behind a Tier 1 release. A single NIST-aligned testing record can support a Texas safe harbor claim, a California anti-bias defense and an ISO 42001 audit at the same time, provided it states thresholds for accuracy, disparity, robustness and security. A test nobody outside engineering reads does not count. The artifact is a signed pre-deployment test result for each Tier 1 system.
Include shadow AI and vendor features embedded in SaaS tools. Record purpose, model and version, data sources, users affected, jurisdictions and decision role, so every later control has a system to attach to.
Score the severity of the affected decision against the degree of human review. Apply the full control set to Tier 1 first, and re-tier on any change to model, data, decision role or jurisdiction.
Use the crosswalk to build one evidence package per control, so one test record serves several regimes. Verify subcategory and clause references against the current published texts before an external audit.
Rate each control from 0 to 3. A total below 15 means the program cannot yet produce defensible evidence for Tier 1 systems, so direct spend at the lowest-scoring controls.
It is one control framework that links ethics principles, legal obligations, risk standards and operational controls. Every control in the pipeline maps to the principle it serves, the laws it satisfies and the NIST AI RMF or ISO/IEC 42001 clause it implements.
That lets one piece of evidence, such as a bias test, support several regulatory requirements at once instead of each statute getting its own program.
A four-layer model, a US and EU legal map as of November 2026, and the tools to apply it. You get a control crosswalk to NIST AI RMF, AI 600-1, ISO/IEC 42001 and ISO/IEC 23894, a risk-tiering matrix by decision criticality, a 14-step implementation checklist and a 0 to 3 scorecard across ten controls.
It covers legal texts, standards and published surveys, and stops there. Legal status reflects sources available as of November 11, 2026, and items such as Colorado rulemaking and federal preemption litigation are still moving.
It is not legal advice and not a study of any company's program. Verify crosswalk references against current published texts before an external audit.
No comprehensive federal AI statute exists, so obligations come from several places. Companies face FTC deception rules, Title VII and ECOA, plus Texas TRAIGA, California's FEHA automated decision system rules and CPPA ADMT rules, NYC Local Law 144, and Colorado's new ADMT law from January 1, 2027.
Firms placing AI on the EU market or serving EU users also face the EU AI Act.
No. Colorado repealed its 2024 AI Act before it took effect. On May 14, 2026, the governor signed SB 26-189, a narrower automated decision-making technology law focused on notice, adverse-outcome disclosures, human review and records.
It takes effect January 1, 2027, if the attorney general completes rulemaking, and penalties reach $20,000 per violation.
NIST defines what to assess and ISO/IEC 42001 runs and certifies the program. The NIST AI RMF provides a risk vocabulary through its GOVERN, MAP, MEASURE and MANAGE functions. ISO/IEC 42001 is a certifiable management system with required clauses and Annex A controls, and it shares its clause structure with ISO/IEC 27001, so existing audit work can be reused.
Not before you know which controls are weakest. Gartner found organizations with AI governance platforms were 3.4 times more likely to report highly effective AI governance, and forecasts AI governance spending of $492 million in 2026.
Score your ten controls first, then direct spend at the lowest-scoring ones, starting with Tier 1 systems.
Drop your details and we'll send Integrated AI Governance: Ethics, Law, Risk and Operational Controls straight to your inbox - no spam, unsubscribe anytime.
Bring your current AI inventory, or the list you wish you had, and the laws you believe apply to it. We will work through which controls already leave evidence, which exist only on paper, and where one test record could serve several regimes. SECTION 7 - FAQ - 5 to 8 questions
Book a governance review