North America, checked 2026-11-02
AI regulation in United States
No single federal AI law, and no federal preemption of state law. NIST's AI RMF is the voluntary national reference. Binding rules come from individual states, and that layer is unsettled rather than merely patchy — Colorado repealed and rewrote its AI Act in 2026, and a federal executive order now directs litigation against state AI laws.
Draft one, the first published version
This is the first public release of AIC's regulatory map and it will change. Orientation, not legal advice: read the primary sources before relying on it. Tell us what is wrong or missing and it goes into the next draft.
NIST AI Risk Management Framework
Administered by U.S. National Institute of Standards and Technology.
What it asks of you
Beside each duty, where it meets the AIC standard. That pairing is AIC's reading, not legal advice.
No binding federal AI statute. Congress has not preempted state AI law, and previous attempts to do so have failed.
Nothing in the AIC standard maps to this.
The NIST AI Risk Management Framework is voluntary. It is the de facto national reference and is what most US procurement language points at, but nothing makes it mandatory.
Binding obligations come from states, and the state picture is actively unsettled rather than merely fragmented — the leading example was repealed and rewritten in 2026.
Sector regulators (FTC, EEOC, financial and health regulators) apply existing law to AI, which is where most live enforcement risk actually sits.
Timeline
- 14 May 2026
- Colorado repealed its original AI Act (SB 24-205) and replaced it with SB 26-189, dropping mandatory risk-management programmes, impact assessments and the standalone anti-discrimination duty.
- 1 Jan 2027
- Colorado's replacement law takes effect — consumer notice for consequential automated decisions, adverse-outcome disclosure within 30 days, and three-year record retention. Enforcement depends on Attorney General rulemaking, currently subject to a court stay.
Who enforces it
State attorneys general for state AI statutes; federal sector regulators under existing authority. A federal executive order has directed the Justice Department to challenge state AI laws and threatened federal broadband funding over them, so the preemption question is live and unresolved.
Every AIC framework that applies in United States
AIC offers three kinds of framework, and they promise different things. Tracking one in the platform maps your evidence to it. An industry framework maps AI onto a safety discipline your sector already uses. Certification is the independent audit.
Track in the platform
Evidence from your connected systems, mapped to each requirement
United States’s own
- NIST AI RMFUS National Institute of Standards and Technology
- NIST CSF 2.0US National Institute of Standards and Technology
- US Data PrivacyCalifornia CCPA and CPRA, including the automated decision rules
- SOC 2AICPA, United States
- HIPAAUS health information
- HITRUST CSFHITRUST Alliance, United States
- CRI ProfileCyber Risk Institute, financial sector
- CMMC Level 1US Department of Defense suppliers
- FedRAMPUS federal cloud services
- CJIS Security PolicyUS criminal justice information
Wherever you operate
- ISO/IEC 42001International standard for AI management systems
- ISO/IEC 27001International standard for information security
Most of these are sector or customer requirements rather than law that binds everyone. Track the ones your customers and regulators ask for.
A mapping says this evidence usually supports this requirement. It is not an auditor's conclusion, and tracking a framework has no bearing on AIC certification. See everything the platform tracks
Industry frameworks
AI mapped onto the safety discipline your sector already runs on
- Engineering / Process Industry, against IEC 61508 / IEC 61511
IEC 61508 and IEC 61511 are international standards, so this mapping applies in United States exactly as it does anywhere else.
- Financial Services, against SR 11-7, Basel III, Sarbanes-Oxley
SR 11-7 is a US supervisory guideline, so in the United States this is the home framework, not a reference point.
- Medical Devices & Health Software, against IEC 62304, FDA GMLP & PCCP
IEC 62304 is international, and the GMLP principles were published jointly by the FDA, Health Canada and the MHRA, so the US FDA is one of their authors.
AIC certification
The independent audit
The standard is the same in United States as everywhere else. What changes is how much of it applies, which depends on how much human judgement sits between your AI and its decisions. See what it would ask of you
3 of the 4 duties above meet one or more of its requirements.
What AIC certification would ask of you here
NIST AI Risk Management Framework is what the law says. This is what the AIC standard says — a separate question, and one that does not change at the border. What changes is which Division your organisation is in, and that depends on how much human judgement sits between your AI and the decision.
“AI decides. Humans review patterns and cases.”
AI makes operational decisions; humans conduct periodic reviews and investigate flagged cases.
Where this shows up
Lender with automated credit decisions + compliance officer reviewing weekly flags. HR tech platform auto-screening applications + recruiter reviewing rejected candidates weekly.
The ones that are hard to fake
- HU-7Override events are evidenced during the assessment period. Zero overrides across material decision volume is not a pass — it triggers interrogation.
- HU-11The Accountable Person can describe operational reality unaided. Needing to check with someone else is a finding.
- EX-5The stated reasons materially match the actual drivers of the decision. Divergence between the reason given and the dominant feature is a critical finding — it is post-hoc rationalisation, not explanation.
- EM-1Adverse automated communications score at least 60 on the AIC Empathy Rubric. A score below 40 blocks certification outright.
- EM-7The disparate impact ratio is at least 0.8 across tested protected characteristics. Refusal to supply the data is recorded as Not Testable, which is itself a finding.
- CO-6The overturn rate is non-zero and explicable. A pipeline where nothing is ever overturned is not a correction pipeline; a very high rate indicates a problem in the underlying model.
- TR-3Disclosure occurs before the interaction affects the person. Post-hoc disclosure fails, and a terms-and-conditions checkbox at signup generally fails for a decision made later.
Measured by: Periodic human review rate is the primary KPI.
Standard v1, issued 2026-09-04. Read all 44 requirements.
AIC certification is not United States compliance and does not establish it. A certified organisation has been assessed against the standard above; whether it satisfies NIST AI Risk Management Framework is a separate question for U.S. National Institute of Standards and Technology, and neither answer substitutes for the other.
This page is a general orientation guide built from public sources, not legal advice, and it does not establish that any organisation complies with United States's requirements. AIC certifies governance against its own published standard; that is a different question from legal compliance, and neither substitutes for the other. Back to the map.