Framework brief · Certificate
ISO 42001 governs the system around the model.
ISO/IEC 42001:2023 is the certifiable management-system standard for AI. It does not make a model safe. It asks whether the company has a system for deciding, and proving, how AI is built and used.
Current as of November 2026. Operator brief, not legal advice.
All six frameworksWhat it is
An AI management system you can be audited against.
ISO/IEC 42001 applies to organizations that develop AI, provide AI, or use AI. The thing certified is the management system in a defined scope — policies, roles, impact assessment, data for AI, the lifecycle from design to retirement, oversight of third-party models, and a way to improve when an impact shows up.
It uses the same management-system shape as other ISO standards. A company that already runs a security management system will recognize the audit rhythm: scope, operate, evidence, surveillance. The scope is different. This one is about AI, not about whether the service’s access reviews happened. HardStack does not claim ISO 27001, and a cousin standard is not a logo we wear.
A certificate with no scope is a slogan.
When it shows up
Buyers started asking for a system, not a principle.
It shows up when the product is the model, or when a regulated buyer has to explain to their own board how a vendor’s AI is governed. Healthcare, finance, and public-sector buyers are the early pressure. A startup that wants the question answered before the RFP arrives is the other.
NIST’s AI Risk Management Framework is the voluntary framework many US teams already speak. ISO 42001 is the one you can be certified against. Buyers sometimes ask for both. They are not substitutes. One is a way to think about risk. The other is a management system an accredited body can audit.
The EU AI Act is law. Obligations depend on what the system does and how it is classified. A 42001 certificate may be useful evidence of governance. It is not a finding that you comply with the Act. Do not let a slide say otherwise.
What done looks like
A certificate from an accredited body, naming the scope. Then surveillance. The certificate does not cover the model you ship next year unless the scope, and the system, actually include it.
Auditors look for the system operating. An AI policy that nobody can connect to a release decision will not survive contact. Impact assessments that are not tied to a real system will not either. The artifact is the trail from policy to the model in production, including models you bought rather than trained.
What it does not replace
- SOC 2, if the buyer is asking whether the service is operated securely.
- HIPAA, if the model touches protected health information. An impact assessment is not a Security Rule risk analysis, though a serious team will connect them.
- GDPR, if personal data is trained on, prompted with, or inferred. Lawful basis and data-subject rights are not annex controls you can waive.
- PCI-DSS, if cardholder data enters the prompt, the log, or the fine-tune.
- FedRAMP, if a federal agency is the buyer of the cloud service. Governing the model is not an authorization to operate.
What the company has to be able to show
Which AI systems are in scope, including third-party ones. Who is accountable. How impact is assessed before a release, and how a bad impact is handled after. What data the system was allowed to see, and what it was forbidden to retain. How a person overrides the system. How a supplier’s model is accepted, monitored, and removed.
If the honest answer is “engineering has a notebook,” you do not have a management system yet. You have material for one. That is a useful place to start. It is not a certificate.
Where Carbon stops
A cue, not a certification body.
ISO 42001 is a framework cue, not a claim that HardStack runs a certified AI management system. Carbon can shape the documentation package — impact, requirements, risk, traceability — to that cue, in your environment. An accredited body still has to audit the system you operate. We do not grant the certificate.
Request a trialRelated briefs