Start With the Purpose, Not the Framework
A framework and a certification are not the same thing. A framework tells you what to do. A certification or assurance program tells someone else (a client for example) that you did it, and that an independent party verified it. Most of the confusion comes from mixing those up, so I find it easiest to sort everything by purpose:
| Category | Purpose | Examples |
| Control frameworks | Control Implementation | NIST AI RMF, ISO/IEC 23894 |
| Management system certifications | Independent Assessment of Governance Program | ISO/IEC 42001 |
| Technical assurance | Assessment of AI Systems / Agents against relevant threats | AIUC-1, HITRUST AI Security Certification |
| Regulatory requirements | Legal requirements | EU AI Act, applicable U.S. state and local laws |
Here's what each category means in practice.
Control Frameworks: What Should We Implement?
These give you the structure, language and controls to build an AI risk program. They're guidance, not a stamp of approval. You can follow them, but nobody certifies you against them.
NIST AI Risk Management Framework (AI RMF)
A voluntary framework released by NIST, organized around four functions: Govern, Map, Measure and Manage. It gives security, legal, data science and business teams a shared way to talk about AI risk.
ISO/IEC 23894:2023
ISO's guidance on managing AI-specific risk. It applies the familiar ISO 31000 risk management approach to AI, so it's a natural fit if your enterprise risk program already speaks ISO.
Management System Certifications: Is Our Governance Independently Assessed?
These look at how your organization governs AI as a whole, not at any single product.
ISO/IEC 42001:2023
The international standard for an AI Management System (AIMS). If you know ISO 27001, the structure will feel familiar: leadership commitment, defined roles, risk and impact assessments, internal audit and continual improvement. An accredited third party can certify it, and ISO/IEC 42006:2025 now sets the requirements for the bodies that audit and certify against it. This is the answer when a customer or your board asks, "How do you govern AI across the company?"
Technical Assurance: Has This Specific System Been Assessed?
An ISO 42001 certificate tells a buyer you have a governed AI program. It doesn't tell them whether the specific system/ agent they're about to connect to their environment holds up against prompt injection. That's the gap these fill:
AIUC-1
A standard built specifically for AI agents, covering data and privacy, security, safety, reliability, accountability and societal risk. Certification combines technical testing, including red teaming, with an audit by an accredited auditor, and the certificate is valid for one year. Maintaining it takes more than an annual recertification: it requires technical retesting every quarter.
HITRUST AI Security Certification
Adds up to 44 AI-specific security requirements on top of a HITRUST e1, i1 or r2 assessment. It's designed for organizations that provide AI systems, not ones that only use someone else's AI tool, and it focuses on security rather than ethics or bias.
These are also useful for third-party risk management. Asking a vendor for an AIUC-1 report or a HITRUST AI certification can tell you a lot more than another custom AI questionnaire.
Regulatory Requirements: What Is Legally Required?
Most frameworks are voluntary. Laws aren't, and this is the area that's changing the fastest. Here's what's in effect today.
EU AI Act
A risk-based law that can apply to organizations outside the EU if they place AI systems on the EU market or their AI output is used there.
U.S. state and local laws
There's no comprehensive federal AI law, so most obligations come from states and cities. These are in effect now:
| Law | Purpose |
| Texas – Responsible Artificial Intelligence Governance Act (TRAIGA) | Sets rules for how AI is developed and deployed in Texas, and is enforced by the Texas Attorney General. |
| California – SB 53, Transparency in Frontier Artificial Intelligence Act | Requires large frontier model developers to be transparent about how they manage the risks of their most advanced models. |
| Illinois – HB 3773 (amends the Illinois Human Rights Act) | Prohibits discriminatory use of AI in employment decisions and requires employers to notify employees. |
| Utah – Artificial Intelligence Policy Act | Requires disclosure when consumers interact with generative AI. |
| New York City – Local Law 144 | Requires bias audits and candidate notices for automated employment decision tools. |
What Does This Look Like in Practice?
Imagine a SaaS company building an AI-powered customer support agent.
The company might use NIST AI RMF to establish a consistent risk assessment methodology, pursue ISO 42001 to demonstrate enterprise-wide AI governance, and use AIUC-1 to provide customers with assurance over the specific agent. Meanwhile, its legal team would evaluate applicable AI transparency, privacy, and consumer protection obligations.
Same organization. Different needs. Different tools.
How to Pick an AI Governance Framework and Where to Start
You don't need to adopt everything on this list. Here's how I'd approach it:
Inventory Your AI
List the AI you build, buy and embed in your products. You can't govern what you haven't found.
Start With Your Legal Obligations
Look at where you operate and how you use AI. Hiring, lending, insurance, healthcare and consumer-facing use cases are where most laws land.
Pick One Control Framework
Choose NIST AI RMF or ISO 23894 as the structure for your program.
Certify Governance If Customers Ask
If customers or your board want independent proof of how you govern AI, ISO 42001 is the way to go.
Add Technical Assurance Where Customers Need Proof
If you sell AI-enabled products or agents, look at AIUC-1 or HITRUST based on what your customers and industry recognize.
Map Once, Comply Many Times
Use crosswalks. AIUC-1 and HITRUST map to other frameworks, so one well-designed control set can support several requirements.
The Bottom Line
AI frameworks aren't necessarily competing with each other. Many are complementary, but they also overlap. The goal isn't to adopt as many as possible. It's to select the right combination based on your risks, regulatory obligations, and assurance needs.
So the next time someone shares a new AI governance framework with you, the question isn't "Do we really need another one?" It's "Which gap does this one fill for us?"