Define the AI system materiality mechanism

In plain terms

Decide what makes a system significant enough to govern formally. Materiality is the gate that decides whether AI9GM applies to a system at all.

What is being judged

Which dimensions of consequence count, and where each one’s line sits. The framework publishes eleven dimensions, any one sufficient: legal and regulatory, individual position, safety, employee, fairness, fundamental rights, privacy, autonomy, financial, operational, reputational.

Two properties matter more than the dimension list.

It is a gate, not a scale. A system tripping one dimension is material. It does not become more material by tripping four, and nothing averages them. The instinct is to score, and scoring produces a system that fails one dimension badly and passes the average.

It has to be applicable without the Board. A mechanism a Business Accountable Executive cannot apply at L4-CLS-03 is a mechanism that routes everything to committee, which is the outcome proportionality exists to prevent.

What this decision does not cover

It does not classify any system; that is L4-CLS-04 and L4-CLS-05. It does not set consequence class, which is severity within the material range.

When it fires

On event. On regulatory change introducing an obligation the current dimensions do not reach. When a system that should have been governed was not, and the mechanism is why. When a dimension is found to be missing, which is how the employee, safety, fairness and reputational dimensions were added.

On cycle. Annually.

What you need before deciding

The regulatory obligation register. The current AI system register, to test the mechanism against systems that already exist. Any system where the determination was contested, which is where the mechanism is unclear.

How this goes wrong

Scoring the gate, above. The dimension nobody owns: a dimension published with no role that would notice it applies, which is how employee impact goes undetected in an organization with no talent representation in the determination. A mechanism only its author can apply: dimensions phrased so abstractly that two people reach different determinations on the same system.

Related decisions

Downstream L4-CLS-03, L4-CLS-04, L4-CLS-05, L5-05, L5-06, L6-03, and the artifact scoping rule at §0.1.

Parallel §5B consequence classification, which grades within the gate this decision sets.

Instrument references

EU AI Act Annex III classifies use cases rather than consequences; a system outside Annex III can be material here. ISO/IEC 42001 clause 6.1. NIST AI RMF MAP 5, impact characterization.