Most organizations already use artificial intelligence — in their products, in internal tools and in everyday work with AI assistants. Far fewer have decided who is accountable for it, which risks are acceptable and how its use is controlled.
AI is already in the organization
In a typical software or engineering company of around a hundred people, AI appears in three places at once. Developers build machine learning features into products. Teams use AI assistants to write, summarize and analyze. And some processes quietly depend on models that nobody formally owns.
None of this is a problem in itself. The problem is that decisions about AI are made one at a time, by individuals, without a shared set of rules. When a client, an auditor or a regulator asks how AI is controlled, the honest answer is often: it depends on who you ask.
What ISO/IEC 42001 actually is
ISO/IEC 42001 is the international standard for an AI Management System (AIMS). Like ISO 9001 for quality or ISO/IEC 27001 for information security, it does not prescribe a technology. It defines how an organization establishes policy, assigns responsibility, assesses risk and impact, controls the AI system lifecycle, and continually improves.
For organizations already certified to ISO 9001 or ISO/IEC 27001, the structure will feel familiar. Much of the management framework — document control, internal audit, management review — can be shared rather than duplicated.
Governance first, certification later
Certification is a reasonable goal, but it should not be the starting point. Many organizations gain most of the value long before an external audit: clear ownership, a register of AI systems, consistent risk assessment and agreed rules for using AI tools.
A practical sequence looks like this:
- Inventory — which AI systems, AI-enabled products and AI tools are in use, and who owns them
- Gap analysis — where current practice stands against the requirements of the standard
- Policy and roles — an AI policy approved by management and clear accountability
- Risk and impact assessment — one method, applied consistently to every relevant system
- Lifecycle controls — how AI systems are designed, validated, deployed, monitored and retired
- Training — people who use AI understand what to share, how to verify results and how to report issues
Only when these work in daily practice does a certification audit make sense. The system is then audited as it is, not as it was written for the occasion.
The regulatory context
In Europe, the EU AI Act introduces obligations that phase in over several years, including requirements for AI literacy among staff who use AI systems. ISO/IEC 42001 is not a substitute for legal compliance, but an AI Management System gives an organization the structure through which regulatory requirements can be identified, assigned and evidenced. For companies outside the EU that serve EU clients, the same expectations increasingly arrive through contracts and tenders.
Proportionate, not bureaucratic
The most common mistake is to build an AI Management System that looks like one for a multinational. A mid-sized company needs documentation that is short enough to be read, processes that fit existing ways of working, and controls that match the actual risk of its AI use. An internal assistant that drafts emails does not need the same treatment as a model that influences decisions about people or public finances.
Proportionality is not a shortcut. It is what makes the system sustainable after the consultants have left.
- AI governance is a management decision before it is a technical one.
- Start with an inventory and clear ownership; certification can follow.
- Integrate with existing management systems instead of duplicating them.
- Scale documentation and controls to the real risk of each AI system.