What does the EU AI Act require of human oversight: proving it is real, not symbolic
A person is named as overseer of a high-risk AI system, and no test shows that the person can monitor, override or stop it. Art. 14 requires the provider to design the system so that assigned persons can understand, monitor, interpret, override and stop it; Art. 26(2) requires the deployer to assign oversight to persons with the authority to act, from the Annex III application date Deferred. This page gives the verification sequence, the records it leaves and the consequence of an untested measure.
Where this starts
The system is high-risk under Art. 6(2), for example a recruitment screening tool under Annex III point 4 or a credit scoring model under point 5(b), and an organisation chart names its overseer. Two roles share the obligation. The provider under Art. 3(3) designs the oversight measures under Art. 14(3): measures built into the system under Art. 14(3)(a) and measures for the deployer to implement under Art. 14(3)(b), both identified before the system is placed on the market and the second set stated in the instructions for use under Art. 13(3)(d). The deployer under Art. 3(4) uses the system in accordance with those instructions under Art. 26(1), assigns oversight to natural persons with the necessary competence, training, authority and support under Art. 26(2), and under Art. 26(3) keeps the freedom to organise its own resources to implement the measures the provider indicated. The deployer hub and the Annex III provider hub carry the obligation set with its dates for each role.
Art. 14(4) states what the assigned person can do, in five capabilities: understand the capacities and limitations and monitor operation under point (a), stay aware of automation bias under point (b), interpret the output under point (c), decide not to use the system or disregard, override or reverse its output under point (d), and intervene or stop the system in a safe state under point (e). Each capability has an observable result on the tested system, and the Art. 14 obligation record lists seven testable elements against them. An oversight measure that exists in a document and produces no effect in a test is the pattern the common findings of both obligation records describe: an override changes the displayed result while the recorded decision still carries the system output; a stop halts the model call while downstream automation keeps running; the stop or override function exists in the provider's product and is out of reach of the assigned overseer in the deployed configuration; the interface pre-fills the output as the decision and no measure counters automation bias; oversight is assigned on an organisation chart with no record of the competence, training or authority of the named persons. The test checklist generator lists the Art. 14 and Art. 26 elements for a given profile.
Regulation (EU) 2024/1689 sets no numeric threshold for any of the five capabilities; the one count in Art. 14 is the two natural persons who separately verify a remote biometric identification under Art. 14(5). No harmonised standard is cited in the Official Journal as of the anchor date, so no presumption of conformity is available for an oversight design. The JTC 21 draft prEN 18229-3Draft fixes the categories of oversight measures and has no public text. prEN 18229-1Enquiry, under enquiry, supplies the log event that records an override, a reversal, a stop or a suspension, which is where the deployer's log duty under Art. 26(6) meets the oversight test. What counts as an effective override or a safe state is the provider's justified choice under Art. 14(3), recorded against the risks, the level of autonomy and the context of use; the rationale step of PR-14-OVR v1.0 captures that record, and the stage of each draft is on the standards page.
The sequence
The work runs in the order below: the measure inventory and the rationale first, then the deployment records, then the scripted campaign, then the log match and the reports.
List each oversight measure in the instructions for use and Annex IV section 2(e), mapped to an Art. 14(4) capability.
Record Oversight measure inventory: one row per measure with type and Art. 14(4) capability
[Regulation (EU) 2024/1689, Art. 14(3)(a) and (b) and Art. 13(3)(d); PR-14-OVR steps 2 to 4]
Record the provider's rationale for each measure and the declared safe state after a stop.
Record Oversight measure rationale record (Art. 14(3)) and safe-state definition (Art. 14(4)(e))
[Regulation (EU) 2024/1689, Art. 14(3) and Art. 14(4)(e); PR-14-OVR steps 5 to 7]
Obtain the deployer's oversight assignment record and compare the deployed configuration with the instructions for use.
Record Human oversight assignment record (Art. 26(2)) and instructions-for-use conformance record (PR-14-OVR output)
[Regulation (EU) 2024/1689, Art. 26(1) and 26(2); PR-14-OVR steps 8 to 9]
Write one scripted test case per measure and provision one assigned account and one unassigned account.
Record Oversight test-case set, log capture configuration and test account register
[prEN 18229-1, enquiry stage, clause on logged event content; PR-14-OVR steps 10 to 12]
Inject a scripted anomaly and record its visibility, the pre-fill state of the decision field and the interpretation result.
Record Monitoring result, pre-fill observation, automation bias verification row and interpretation result per case
[Regulation (EU) 2024/1689, Art. 14(4)(a) to (c); PR-14-OVR steps 14 to 19]
Execute not-use, disregard, override and reverse from the assigned account and read the recorded decision for each case.
Record Override case records and recorded-decision comparison per case
[Regulation (EU) 2024/1689, Art. 14(4)(d) and Art. 26(2); PR-14-OVR steps 20 to 22]
Activate the stop and the suspension procedure, record the post-stop state, then repeat one stop from the unassigned account.
Record Stop case record, post-stop state record, authority negative-test record and suspension case record
[Regulation (EU) 2024/1689, Art. 14(4)(e), Art. 26(2) and 26(5); PR-14-OVR steps 23 to 26]
Extract the log event for every oversight action and match each event to its test case by time and actor.
Record Oversight action log extract (Art. 14(4)(d) and (e), read with Art. 12) and event-to-case match table
[Regulation (EU) 2024/1689, Art. 12(2)(c) and Art. 26(6); PR-14-OVR steps 30 and 35; PR-12-LOG step 21]
Write the verification reports with the finding register classified by the Art. 14 or Art. 26 element.
Record Human oversight verification report and override, reversal and stop verification report (PR-14-OVR output)
[PR-14-OVR steps 38 to 43]
Evidence this situation ends with
| Artifact | Answers | Produced by |
|---|---|---|
| Oversight measure rationale record linking each measure to risk, level of autonomy and context of use (Art. 14(3)) | Art. 14(3): each measure named with the risk, the level of autonomy and the context of use it answers | the provider |
| Human oversight assignment record (Art. 26(2)) | Art. 26(2): the natural persons who hold oversight, with their competence, training and authority recorded | the deployer |
| Instructions-for-use conformance record (PR-14-OVR output) | Art. 26(1): the deployed configuration, input types and use cases fall within the intended purpose and conditions of use | PR-14-OVR v1.0 |
| Override, reversal and stop verification report (PR-14-OVR output) | Art. 14(4)(d) and (e): the recorded decision carries the person's choice, and the stop brings the system to the declared safe state | PR-14-OVR v1.0 |
| Automation bias measure description and verification record (PR-14-OVR output) | Art. 14(4)(b): the output is not pre-filled as the decision, and the assigned person records an active choice | PR-14-OVR v1.0 |
| Safe-state definition and post-stop state record (Art. 14(4)(e)) | Art. 14(4)(e): the declared safe state, including downstream automation, against the state observed after each stop | PR-14-OVR v1.0 |
| Human oversight verification report (PR-14-OVR output) | Art. 26(2), (3) and (5): each deployer-side measure implemented and exercisable, a stop or override from the assigned account takes effect, and the suspension procedure halts use | PR-14-OVR v1.0 |
| Oversight action log extract showing override and stop events (Art. 14(4)(d) and (e), read with Art. 12) | Art. 12(2)(c) and Art. 26(6): each override, reversal, stop and suspension as a timestamped event with the actor identity, matched to its test case | PR-14-OVR v1.0, input to PR-12-LOG v1.0 |
Each artifact follows a specification with a synthetic sample on the deliverables page: the verification reports, the state records and the log extract follow the test report specification, and the rationale record and the assignment record are inputs the provider or the deployer holds in the file before the campaign runs.
If nothing is done
A high-risk system in use after the Annex III application date whose oversight measures are described in the technical documentation and produce no effect in a test is subject to the market surveillance procedure of Art. 79: the authority evaluates the system and, where the requirements of Chapter III are not met, requires the provider under Art. 79(2) to take corrective action, to withdraw the system or to recall it within a period the authority sets. A finding against the provider obligations in Art. 16, which include the Art. 14 design, falls in the penalty tier of Art. 99(4)(a), and a finding against the deployer obligations in Art. 26 falls in the tier of Art. 99(4)(e); the amount in a given case is set by the national authority under the Art. 99(7) criteria. On the deployer's side, Art. 26(5) requires a deployer with reason to consider that use in accordance with the instructions may present a risk within the meaning of Art. 79(1) to inform the provider and the market surveillance authority and to suspend use, and a suspension procedure that has never been exercised is the measure the deployer relies on at that moment. The sequence above produces the records a provider supplies on a reasoned request under Art. 21(1) and a deployer supplies in the cooperation with competent authorities that Art. 26(12) requires.
Services, guides and tools for this situation
The logging and human-oversight verification service runs the override, stop and suspension campaign under PR-14-OVR v1.0 and the log extract match under PR-12-LOG v1.0, within the limits stated on what we cannot do.
- Guides
- ICP hubs
Every inquiry receives a reply within 24 hours. Scope and availability are discussed per request.