Monitoring, Incidents & Disclosure — Visibility
Part of EU AI Act — Structural Alignment and Structural Alignment in Practice. A structural reading, not legal advice.
The EU AI Act asks providers of high-risk AI systems to monitor those systems after they are placed on the market, and to report serious incidents to the authorities. In defined cases it also requires that people be told when they are interacting with an AI system or being shown AI-generated content.
Each of these makes a system's behavior observable to someone: the provider, the authority, the person affected. A structural reading asks whether observable is the same as visible.
From Observation to Visibility
Post-market monitoring collects and analyses data on how a system performs after deployment. That shows what the system is doing in aggregate. It does not, by itself, show where any single determination was made.
Visibility, in the structural sense, is narrower and harder: a determination made in autonomous execution can be traced to where and how it was made. A monitoring system can be comprehensive and still be unable to say which component produced which effect, on whose authority.
Untraceable Determination
As systems delegate across agents, tools and workflows, determinations increasingly occur with no human decision point. Monitoring sees outcomes and patterns. The determination itself happens inside the execution.
Where nothing marks that moment, responsibility has nothing to attach to. This is how a system can appear operationally successful while its reasoning has already gone dark, the condition described as Shadow AI. The Responsibility Boundary matters here: a determination is only accountable if the system can show that it crossed one.
Incident Reporting Depends on Visibility
A serious-incident report asks the provider to establish what happened and why. If determinations were never visible in operation, the report is a reconstruction: an account written afterwards from partial evidence.
The two structural questions are distinct. Decision lineage preserves why a decision was made. Visibility ensures that a determination was seen to be made at all. Once divergence has been detected, audit and repair take over; see Risk Management — Audit and Repair.
Disclosure Is Visibility Toward Persons
Telling people they are dealing with an AI system is visibility at its smallest scale: a person can recognize which interactions are machine-made. It is necessary, but it is not sufficient. Disclosing that a system is artificial says nothing about where inside it a determination was made.
Continuity Under Acceleration
As autonomy and speed increase, determinations outnumber human reviews. Visibility cannot then be a periodic sample. It has to be a property of execution.
It also underpins Continuity of Agency. A person cannot keep a current understanding of a system whose determinations they cannot see.
Closing Reflection
The AI Act requires monitoring, reporting and disclosure. Structural alignment asks: Can the system show where its determinations were made?
If not, visibility remains observation of outcomes. If so, it becomes traceability of execution. Audited State proposes visibility as the continuity-preserving interpretation of monitoring and disclosure obligations under acceleration.