Free tool · Article 14
EU AI Act Article 14 Agent Audit Gap Checker
Paste your agent architecture JSON below — or tick the controls you already have — and instantly map missing human-oversight and audit-trail controls against EU AI Act Article 14. Everything runs in your browser; nothing you paste leaves your device.
1. Paste your architecture JSON
Include keys for the controls you have — for example human_oversight, kill_switch, audit_log, override, anomaly_detection, operator_training, explainability, and oversight_roles. The checker searches keys and values, case-insensitively.
That JSON could not be parsed. Fix it, or use the checklist below instead — the checklist works on its own.
2. Or confirm controls manually
Tick every control that is actually implemented and tested — not just planned. Manual ticks are combined with anything found in your JSON.
Your gap map
This is an informational self-assessment, not legal advice. It maps the controls described in Article 14 — it does not determine your legal classification under the EU AI Act. For binding answers, consult qualified counsel.
What Article 14 actually requires
Article 14 requires high-risk AI systems to be designed so they can be effectively overseen by natural persons while in use. Oversight must be proportionate to the risks, the level of autonomy, and the context of use. The people assigned to oversight must be able to understand the system's capacities and limitations, monitor its operation and detect anomalies, stay aware of automation bias, correctly interpret its output, and — critically — decide not to use it, disregard or override its output, and intervene or interrupt it, including through a stop procedure.
- Understandability (Art. 14(4)(a)): overseers understand capacities, limitations, and risks.
- Monitoring & anomaly detection (Art. 14(4)(b)): operation is watched; dysfunctions and unexpected performance are detected.
- Automation-bias awareness (Art. 14(4)(c)): overseers are trained not to over-rely on the output.
- Interpretation (Art. 14(4)(d)): output can be correctly interpreted, using the tools and information provided.
- Override & interruption (Art. 14(4)(e)): overseers can disregard, override, or reverse output and stop the system safely.
An audit trail is what proves any of this happened. Without timestamped records of who oversaw what, which output was overridden, and when the system was stopped, oversight exists on paper only.
FAQ
Is this legal advice?
No. This tool is an informational self-assessment for operational readiness — it is not legal advice and does not determine your legal classification under the EU AI Act.
Does my pasted JSON leave my device?
No. The analysis runs entirely in your browser with JavaScript. Nothing you paste is sent to a server.
What if I don't have an architecture JSON?
Use the manual checklist in step 2. It covers the same eight controls and produces the same gap map.
What counts as a gap?
Any of the eight Article 14 oversight/audit controls that is neither found in your JSON nor ticked in the checklist. Each gap is listed with the Article 14 duty it maps to and the concrete fix.
Close the gaps, don't just list them
The Governed Agent Mesh Playbook — Studio Edition ($29) gives you the approval-gate patterns, audit-trail templates, override procedures, and incident runbook behind every fix on your list, ready to run.
Get the Studio Edition — $29