Home / Services / AI Governance & ISO 42001
AI Governance & ISO 42001
ISO/IEC 42001 is the international standard for AI management systems. We help organisations find the AI they are actually running, understand the risk it carries, and build governance that holds up — technically, and in front of an auditor.
Our angle
Security practice, not audit practice
We assess the management system and the attack surface: prompt injection, data poisoning, model extraction, and the way autonomous agents use the permissions you give them.
Most ISO 42001 consultants can do the first. Far fewer can do the second.
What changed in 2026
The EU AI Act’s high-risk obligations were deferred. Standalone high-risk systems now have until 2 December 2027, and AI embedded in regulated products until 2 August 2028. Prohibitions, general-purpose AI rules and the Article 50 transparency obligations already apply — the last of those landed on 2 August 2026 and was never part of the deferral.
That is not a reason to wait. It is the difference between building governance that works and assembling documentation that merely exists.
Digital Omnibus on AI · final Council approval 29 June 2026
What we do
Four steps, each sellable on its own
AI inventory & risk triage
Find the AI actually in use, including the tools business units adopted without telling IT. Classify by risk, and against the AI Act’s categories where they apply.
You get: an AI register, risk classification and a prioritised view · 1–2 weeks
ISO 42001 gap analysis
Assessment against the standard’s management clauses and Annex A controls, with an honest view of the distance to certification.
You get: gap report, remediation plan, certification timeline · 3–5 weeks
AI security assessment
Threat modelling of a specific system: prompt injection, training and retrieval data poisoning, model extraction, and the provenance of the models you rely on.
You get: threat model, ranked findings, control recommendations · per system
Zero Trust for AI agents
Agent identity, permissions scoped per task rather than granted standing, tool access limits, and protection of memory and context against poisoning.
You get: agent security assessment and remediation path · per deployment
Identity, least privilege and segmentation applied to a non-human actor — the disciplines we have worked in for years
Worth being clear about
ISO 42001 certification is not EU AI Act compliance
The two overlap substantially, and certification is strong evidence of governance maturity, but they are different instruments with different scopes. Certification helps you demonstrate control. It does not, by itself, discharge your obligations under the Act.
Anyone who tells you otherwise is selling you something.
Frequently asked
What is ISO/IEC 42001?
The international standard for an AI management system, published in 2023. It sets requirements for establishing, implementing, maintaining and improving the governance of artificial intelligence within an organisation, and it is certifiable by accredited certification bodies.
How long does certification take?
For an organisation already certified to ISO 27001, typically six to twelve months from gap analysis to certification audit. Without an existing management system, longer.
What is Zero Trust for AI agents?
Applying Zero Trust principles — verify explicitly, grant least privilege, assume breach — to autonomous agents. In practice: identities that are cryptographically rooted rather than shared, permissions scoped to a single task, constrained tool access, and protection of memory and context against poisoning.
What is shadow AI, and how do you find it?
AI tools adopted by teams without review by security or IT. It is found through discovery across code repositories, cloud environments and user activity — not by asking people, who routinely do not know or do not think to mention it.
We only use AI, we do not build it. Does this apply?
Yes. Deployers carry obligations under the EU AI Act, and in practice most AI risk arrives through third-party tools adopted without review.
Most engagements begin with the inventory, because most organisations cannot yet answer “what AI are we running?” with confidence. It is a short piece of work, and it usually changes the conversation.