EUR 35 million or 7%
For engaging in a prohibited AI practice.
Article 99The EU AI Act is law. The NIST AI RMF is voluntary. ISO/IEC 42001 is a standard you can certify against. Here is how the three differ, what each asks of you and how one program can meet them all.
AI compliance frameworks are the laws, standards and guidance that set out how organizations should build, buy and run AI responsibly.
The three most enterprises work to are the EU AI Act, a binding regulation; the NIST AI Risk Management Framework, a voluntary US framework; and ISO/IEC 42001, an international management system standard. They overlap heavily, so one inventory and one set of controls can produce the evidence for all three.
One is a law, one is guidance and one is a standard an auditor can certify. Knowing which is which tells you what is required, what is expected and what is optional.
| EU AI Act | NIST AI RMF | ISO/IEC 42001 | |
|---|---|---|---|
| What it is | A binding EU regulation, (EU) 2024/1689 | A voluntary framework from NIST, the US standards agency | An international standard for an AI management system |
| Since | In force since August 1, 2024 | Version 1.0 released January 26, 2023 | Published December 2023 |
| Who it applies to | Providers and deployers of AI systems placed on the market or used in the EU | Any organization that chooses to adopt it | Any organization that provides or uses AI systems |
| How it is built | Risk tiers: prohibited practices, high-risk systems, transparency duties and general-purpose AI | Four functions: Govern, Map, Measure and Manage | Requirements to establish, run and continually improve an AI management system |
| How you show it | Technical documentation and conformity assessment for high-risk systems | Self-assessment. There is no certificate | An independent certification audit |
| Penalties | Fines up to EUR 35 million or 7% of worldwide annual turnover | None. It is voluntary | None. It is voluntary |
The Act applies in stages, and the 2026 AI Omnibus moved the high-risk dates. These are the dates on the European Commission’s implementation timeline. This is a practitioner’s summary, not legal advice.
| Date | What applies |
|---|---|
| August 1, 2024 | The AI Act enters into force |
| February 2, 2025 | Prohibited AI practices and AI literacy obligations apply |
| August 2, 2025 | Rules for general-purpose AI models apply |
| July 27, 2026 | The AI Omnibus amendments enter into force |
| August 2, 2026 | Transparency rules under Article 50 start to apply |
| December 2, 2026 | Transitional deadline for marking AI-generated content from systems already on the market |
| December 2, 2027 | Rules for high-risk AI systems listed in Annex III apply |
| August 2, 2028 | Rules for high-risk AI embedded in regulated products under Annex I apply |
Article 99 sets maximum fines by type of breach. For companies, each cap is the fixed amount or the percentage of worldwide annual turnover, whichever is higher. Lower caps apply to smaller companies.
For engaging in a prohibited AI practice.
Article 99For breaching most other obligations, including those on providers and deployers of high-risk systems and the transparency rules.
Article 99For giving incorrect, incomplete or misleading information to authorities or notified bodies.
Article 99NIST released the AI Risk Management Framework on January 26, 2023, for voluntary use. A Generative AI Profile followed on July 26, 2024, and NIST says a revised version is in progress. Its core is four functions.
Set the policies, roles and accountability for AI risk. This is where the framework calls for mechanisms to inventory AI systems.
NIST AI RMF coreEstablish the context for each AI system: what it is for, who it affects and which risks come with it.
Analyze, assess and track those risks with tests and metrics, before deployment and in operation.
Prioritize the risks you found and act on them, then keep monitoring as the system changes.
ISO/IEC 42001 was published in December 2023. Where the EU AI Act regulates AI systems and the NIST AI RMF guides risk decisions, this standard describes how an organization should manage AI as a whole.
It specifies requirements for establishing, implementing, maintaining and continually improving an AI management system.
ISO/IEC 42001:2023It applies to any organization, whatever its size or type, that provides or uses products or services built on AI systems.
An independent auditor can certify an organization against it, which gives customers and partners a certificate they can check.
The three frameworks ask for the same things in different words: know what AI you have, assess its risk, control it and show your work. Cranium AI runs that as the AI Trust Loop, so one program produces the evidence for each.
Inventory every model, agent and vendor. All three frameworks start here, and none can be met without it.
Watch how AI behaves in production, so risk assessments rest on evidence instead of assumptions.
Write controls once and map them to the EU AI Act, NIST AI RMF and ISO/IEC 42001, alongside your own internal standards.
Test systems against real attacks and enforce guardrails, the robustness and security work the frameworks expect.
Produce current evidence for each framework on demand, for auditors, customers and regulators.
Each of these goes further into one framework or one part of the work.
From high-risk classification to technical documentation, step by step.
Read the postHow to keep NIST AI RMF and ISO 42001 evidence current instead of rebuilding it each audit.
Read the postOne place for controls, owners and evidence across every framework.
Read the postApplying the framework to the AI your third parties bring in.
Read the postPolicy maps to the EU AI Act, NIST AI RMF and ISO 42001 from the start, with your internal standards encoded alongside them.
See GovernCranium ComplianceAgent (Beta) drafts cited answers to framework questions from your own documentation, and on-demand scoring against the NIST AI RMF Starter Pack shows where each system stands.
See AI complianceCranium AI Cards are living transparency reports for each system, mapped to all three frameworks and ready to share.
See ProveA risk-by-risk map to the OWASP Top 10 for LLM and agentic applications sits next to the regulatory frameworks.
See the OWASP mapStraight answers, kept current. Last updated .
Ask us anything →The three most enterprises work to are the EU AI Act, a binding EU regulation; the NIST AI Risk Management Framework, a voluntary US framework; and ISO/IEC 42001, an international standard for an AI management system. Security teams often add the OWASP Top 10 lists for LLM and agentic applications.
No. NIST released the AI Risk Management Framework on January 26, 2023, for voluntary use. There is no certificate and no penalty. Organizations adopt it because it gives customers, partners and regulators a recognized reference.
After the 2026 AI Omnibus, the rules for high-risk AI systems listed in Annex III apply from December 2, 2027, and the rules for high-risk AI embedded in regulated products under Annex I apply from August 2, 2028. Prohibited practices, AI literacy and general-purpose AI obligations already apply.
Article 99 sets three tiers: up to EUR 35 million or 7% of worldwide annual turnover for prohibited practices, up to EUR 15 million or 3% for most other breaches, and up to EUR 7.5 million or 1% for giving incorrect or misleading information to authorities. For companies the higher figure applies, with lower caps for smaller companies.
Yes. ISO/IEC 42001 is a management system standard, so an independent auditor can certify an organization against it. The EU AI Act and the NIST AI RMF do not offer an equivalent organization-wide certificate.
The EU AI Act is law: it sets obligations by risk tier and carries fines. The NIST AI RMF is voluntary guidance: it gives organizations four functions, Govern, Map, Measure and Manage, for handling AI risk. The two fit together: the NIST functions give teams a working method, and the Act sets what must be met.
Cranium ONE maps every AI system to the EU AI Act, NIST AI RMF and ISO 42001, drafts cited framework answers with ComplianceAgent (Beta), scores NIST AI RMF compliance on demand and produces AI Cards as shareable evidence. It runs as the AI Trust Loop: Discover, Observe, Govern, Secure and Prove.
See how Cranium AI discovers, observes, governs, secures and proves every model, agent and vendor in your enterprise.
Cookie Policy · Privacy Policy
Global Privacy Control detected. Opt-out request honored: marketing tags stay off in this browser.