← Back to concept explainers

Concept Fable

Conformity assessment

Conclusion first:Conformity Assessment is the procedure for verifying whether a regulated AI system meets applicable legal and standard requirements before it is placed on the market or put into use. For high-risk AI/Agents, this typically involves technical documentation, risk management, data governance, testing, logging, human oversight, cybersecurity, a quality management system, and, where necessary, third-party assessment. Major changes after launch may also require reassessment.

Conformity Assessment · Regulation/Compliance

Read the story first. There is no rush to give a definition here; let the question grow on its own first.

Fable

Before the red cabinet machine goes onto the bridge, it first passes three bridge-verification gates

The city's newly built automatic bridge gate was about to be commissioned. It would identify vehicles, collect bridge tolls, block dangerous cargo, and send the names of violators to the patrol team. The craftsman said the gate had been tested a hundred times, and the city lord was eager to cut the ribbon. But the old bridge inspector blocked the door: being able to run doesn't mean it's qualified. In the past, when a small tool broke, you just fixed it. But if this bridge gate makes a mistake, it could block an ambulance, wrongly fine a caravan, or leak someone's whereabouts, affecting traffic across the whole city.

The craftsman first submitted a stack of promotional drawings. The bridge inspector returned them and required risk documents, design descriptions, data sources, test results, human oversight plans, record formats, and safety certifications.

Bridge inspection has three gates: whether the documents are complete, whether the controls actually run, and whether the entire installation complies with applicable rules. High-risk parts also require review by an independent body.

One test failed: the misrecognition rate for out-of-town license plates in night rain was too high. After the craftsmen corrected the data and added manual fallback, the bridge gate received its qualified mark.

After launch, a pass mark is not a free pass either. Major changes, incidents, and performance drift can all trigger re-evaluation.

The city thus understood that conformity assessment is not about sticking a medal on a machine, but about proving that a high-risk complete installation meets specified requirements before it enters the market or a critical process.

The bridge inspector also had the patrol team, the clinic, and the chamber of commerce each send one person to try it. The patrol team cared about toll evasion, the clinic cared about emergency vehicles passing, and the chamber of commerce cared about cargo trucks being wrongly stopped. Once all three tried it, they discovered that the same gate posed different risks in different people's eyes.

Only then did the city lord understand that bridge inspection was not slowing down the opening. The more a device was meant to enter the main road, the less it could rely solely on its maker's claim that it could run; it had to lay out its origins, boundaries, failure handling and review methods on the table. Later, whenever a new mechanism went onto the bridge, the craftsmen would first ask: who could this thing harm in the worst case, and who should prove it has been reviewed. The bridge inspector carved this sentence by the gate to remind those who came later to prove before passing. This was also the precondition for the city to be willing to entrust passage to it. Otherwise, a lively ribbon-cutting would only push risk onto the bridge deck.

Reveal

This story is about: conformity assessment

Conformity Assessment is the procedure for verifying whether a regulated AI system meets applicable legal and standard requirements before it is placed on the market or put into use. For high-risk AI/Agents, this typically involves technical documentation, risk management, data governance, testing, logging, human oversight, cybersecurity, a quality management system, and, where necessary, third-party assessment. Major changes after launch may also require reassessment.

What matters is that it is not just a term, but a structural problem that repeatedly appears in real AI / Agent systems. Understanding it lets you judge when to add a model, when to change a process, and when to add governance.

Metaphor mapping

  • Automatic Bridge Gate: Regulated High-Risk AI Systems
  • Promotional image: marketing material insufficient to prove compliance
  • Three bridge inspection gates: document, control and rule compliance checks
  • Independent body review: third-party conformity assessment
  • Night rain license plate failure: a defect found by evaluation
  • Major change re-assessment: ongoing compliance requirements

Andy Ren Xu's judgment

The practical value of this concept is that it helps you break "AI features that look smart" into deliverable, verifiable, and governable units of work.

EU AI ActCE markinghigh-risk AItechnical documentation
京ICP备2026014272号