ISO/IEC 42001:2023 · AI Management System · Certification
FRAMEWORKISO 42001 is the first management-system standard for AI. Uproot runs it against your real model lifecycle impact assessments, data provenance, and eval gates proven, not just claimed.
The first certifiable AI standard
Builds on your ISO 27001 management system
acme-ai · ISO 42001:2023
AIMS · 6 AI systems in scope
79%
cert-ready
Annex A · control objectives
38 controls
AI policy
A.2.1–A.2.4
100%
Impact assessment
A.5.1–A.5.5
82%
AI system life cycle
A.6.1–A.6.2
74%
Data for AI systems
A.7.1–A.7.6
71%
Third-party & suppliers · models
A.10.1–A.10.4
68%
AI systems
6 registered
Cert body
Schellman
Open impacts
2
Impact assessment triggered
A new model hit the registry. A.5.2 impact assessment was opened and routed to its owner.
Annex A controls
38Across nine control objectives covering AI policy, lifecycle, data, transparency, and third parties.
Control objectives
9From AI policy and internal organization through impact, lifecycle, data, and the use of AI systems.
Published
2023The world's first certifiable AI management system standard net-new, not a profile of something older.
Certification cycle
3yearsStage 1 + Stage 2, annual surveillance, recertification at year three like ISO 27001.
Builds on
27001Same management-system backbone. If you run an ISMS, the AIMS slots straight on top.
ISO 42001 turns "responsible AI" from a values statement into a management system: policy, impact assessment, a governed lifecycle, data provenance, and third-party model oversight. Uproot proves each against your real ML stack.
AI policy
A documented AI policy aligned to your objectives and risk appetite the foundation the AIMS hangs from.
Impact assessment
Assess each AI system's impact on individuals and society fairness, safety, consequences before and during deployment.
AI system life cycle
Objectives, design, validation, deployment, and monitoring where the model registry and CI gates earn their keep.
Data for AI systems
Provenance, quality, and governance of training data knowing where it came from and whether it fits.
Third-party & suppliers
Governing foundation models, APIs, and suppliers across your AI supply chain.
Verification and validation is where responsible-AI intentions meet the pipeline. Here's how Uproot turns your eval gates into continuous, certifiable evidence.
AI systems must be verified and validated against their requirements before and during use including performance, robustness, and bias testing with results documented. The control that separates a governed model from a hopeful one.
"When we evaluated our options for compliance and securing our systems, we found that UprootSecurity's compliance and security model aligned perfectly with our needs. It gave our team real-time visibility into the end-to-end process, saving our engineers hundreds of hours of manual effort."
A.6.2.4 · VERIFICATION & VALIDATION
AI systems are verified & validated
Last evidenced 09:14:02 UTC · 6 AI systems · sha256 verified
The control
What Annex A.6.2.4 asks for
Your ML stack
What we read from your systems
Uproot points at the systems that actually gate models the registry, the eval jobs, the deploy pipeline.
Evidence collected
Eval results, gates, and lineage
Eval coverage, pass thresholds, and the link between a deployed model version and its passing run are pulled from source, hashed, and stored.
For the auditor
Read-only portal, mapped to the SoA
The certification body opens A.6.2.4 and sees every in-scope model, its eval history, and proof that the deploy gate is blocking already linked to your Statement of Applicability.
SoA · included · Owner· ML Platform · Nonconformities · 0
A path for a team that already runs ISO 27001 with Uproot. Net-new management systems take a little longer to stand up the clauses.
Week 0
Inventory AI systems
Register every model and AI feature in scope. Uproot discovers them from your model registry and deployment pipeline.
Week 1–2
Set AI policy & scope
AI policy, roles, and AIMS scope defined. The Annex A Statement of Applicability is generated.
Week 3–6
Impact & risk
Impact assessments run per system; AI risks assessed and treated. Data provenance and eval gates wired to evidence.
Week 7–9
Stage 1 audit
The body reviews your AIMS documentation through the read-only portal. Findings closed before Stage 2.
Week 12–13
Stage 2 audit
The body tests that controls operate across the AI lifecycle. Evidence is continuous nothing to assemble.
Year 1–3
Certified + surveillance
Certificate issued. Surveillance audits read the same live evidence. Recert at year three is routine.
An “AI principles” page on the website with nothing operational behind it
Impact assessments done once, in a doc, never revisited when the model changes
No record of which model version is in production or whether it passed its evals
Foundation-model suppliers used with no governance or allocated responsibility
An AI policy tied to operating controls across the real model lifecycle
Impact assessments triggered automatically when a model enters or changes scope
Every deployed model linked to its passing eval run, gates proven blocking
Third-party models inventoried and governed, responsibilities documented
A partial map of the AIMS evidence Uproot pulls from your ML and data stack registries, pipelines, datasets, and suppliers.
Lifecycle & evals
MLflow · model registry
6
GitHub · eval workflow
gated
W&B · eval runs
live
Argo · deploy gate
blocking
Data for AI
Dataset registry
provenance
Snowflake · lineage
tracked
Data quality checks
CI
Consent / licensing
recorded
Impact & risk
Impact assessments
6
AI risk register
live
Bias / fairness tests
enforced
Human-oversight log
on file
Suppliers
Foundation models
3
API supplier DPAs
signed
Model cards on file
linked
Responsibility matrix
live
ISO 42001 is new, and the accredited body pool is growing fast. Uproot is body-agnostic and gives yours a read-only portal scoped to your AIMS the same handoff that works for ISO 27001.
BSI
ukas
Schellman
anab
TÜV SÜD
dakks
DNV
accredia
A-LIGN
anab
Bureau Veritas
ukas
SGS
ukas
Coalfire
anab
A-LIGN AI
anab
Mastermind
anab
NQA
ukas
+ growing
on request
It's the first certifiable management-system standard for artificial intelligence an "AIMS." It governs how you develop, deploy, and oversee AI systems responsibly, the way ISO 27001 governs information security.
Any organization building or meaningfully using AI systems, especially those selling to enterprises or operating in regulated markets. It's quickly becoming the artifact procurement teams ask for alongside SOC 2.
They're complementary. The EU AI Act is law with risk-tiered obligations; ISO 42001 is a voluntary management system that demonstrates many of those practices. Running the AIMS makes Act readiness far easier to evidence.
No, but it helps. ISO 42001 shares the same management-system clauses (4–10). If you already run an ISMS with Uproot, the AIMS reuses that backbone and a lot of the evidence.
An assessment (Annex A.5) of how an AI system affects individuals and society fairness, safety, and consequences. Uproot triggers one whenever a model enters or materially changes scope, and tracks it to closure.
Annex A defines 38 controls across nine control objectives. As with ISO 27001, you justify applicability in a Statement of Applicability which Uproot generates from your real AI inventory.
Register your AI systems, generate impact assessments, and prove your eval gates and data provenance continuously. Invite your certification body when Stage 2 is a formality.
