> ## Documentation Index
> Fetch the complete documentation index at: https://docs.orq.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Human oversight

> Map Orq.ai controls to EU AI Act Article 14 duties, and see where each control stops.

EU AI Act Article 14 requires high-risk AI systems to be built so natural persons can effectively oversee them, with measures proportionate to the risk, the autonomy and the context of use. **Orq.ai** supplies configurable controls that can serve as those measures. It does not make a system compliant.

## Who owes what

* **Provider**: builds the oversight measures into the system before placing it on the market and states them in the instructions for use (Article 14(1), 14(3)(a), 13(3)(d)).
* **Deployer**: assigns oversight to natural persons with the competence, training, authority and support to exercise it (Article 26(2)).

> [card] [Read Article 14](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32024R1689#art_14)
>
> Article 14 of Regulation (EU) 2024/1689, the EU AI Act.

| Article 14 duty | Control | Where it stops |
| - | - | - |
| Understand limits and monitor operation (14(4)(a)) | [Traces](https://docs.orq.ai/ai-studio/observability/traces.md) show the model, tools, tokens, latency and cost of a run; [evaluations](https://docs.orq.ai/ai-studio/observability/trace-evaluations.md) score outputs | Declared limits come from the instructions for use. Nothing flags an anomaly that a rule does not define |
| Stay aware of automation bias (14(4)(b)) | [Annotations](https://docs.orq.ai/ai-studio/observability/annotations.md) record agreement with, or an override of, the machine score | Awareness comes from the instructions and from training, not from a control |
| Interpret an output (14(4)(c)) | Traces show the inputs, retrieved context, tool calls and model behind an output | The interpretation aids are the provider's to specify (13(3)(d)) |
| Decide not to use, disregard, override or reverse an output (14(4)(d)) | Annotations record the decision and keep the evaluator output beside the correction | A blocking guardrail fires before a person sees the output; it is a measure under 14(3)(a), not a human decision |
| Approve or reject a tool call (supports 14(4)(d)) | Tool approval pauses an Agent before a tool runs | Enabled only when `tool_approval_required` is `respect_tool` or `all`; an AI Studio Agent defaults to `none` |
| Intervene or interrupt a run (14(4)(e)) | Nothing on this page | Max. turns and runtime limits are ceilings set before a run |
| Verify an identification with two persons (14(5)) | Not covered | Review is single-reviewer |

## Where the controls stop

* Output guardrails do not run on streaming responses.
* Max. execution time excludes tool execution, and a request value overrides the agent manifest.
* Tool approval is not enforced on `POST /v3/router/responses`, and **Allow always** removes later pauses.
* No control here halts a run already in flight.

## Records

[Audit logs](https://docs.orq.ai/ai-studio/organization/audit-logs.md) record control-plane changes; traces and annotations record per-run actions and human decisions. Those are the logs of Article 12, kept under Article 26(6). A record shows an action was taken, not that oversight was effective.

## Change control

Prompts and model deployments are versioned and audited, so a change ties to the person and time that made it. A change to a deployed high-risk system can be a substantial modification: the party that makes it becomes a provider (Article 25(1)(b)), with the obligations of Article 16 and a new conformity assessment under Article 43(4).

## What Orq.ai does not decide

* The high-risk classification and the acceptance thresholds.
* Who oversees, and with what competence, training, authority and support.
* Whether to suspend use and notify under Article 26(5).

Document those decisions outside the platform and reference the **Orq.ai** traces, annotations and audit entries that evidence them.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.