Guide

AI Act standards: agent control mapping

Map agent authority, access, policy, validation, oversight and evidence to prEN 18228, prEN 18282 and prEN 18229-1 preparation work.

For financial-services platform architects, AI risk leads and security teams.

Last updated: Sep 7, 2026 · Version v1.0 · Not legal advice.

Short answer

Use a control-to-evidence register to prepare agent systems for risk management, cybersecurity and logging assessment. The matrix below is KLA’s topic-level engineering interpretation. Clause-level conformity needs the applicable standard text and a system-specific assessment.

7 September 2026

Verified standards status

The JTC 21 catalogue entries updated 4 September show the following status. The Commission portal reports EN 18286 assessment as ongoing. Earlier enquiry dates are historical.

ReferenceSubjectCatalogue stageOJEU citation recorded
EN 18286:2026Quality managementPublishedNo
prEN 18228Risk managementApprovalNo
prEN 18282CybersecurityApprovalNo
prEN 18229-1LoggingApprovalNo
Preparation matrix

Map a control to an observable result

Treat each row as an engineering work package. The draft identifiers indicate relevant topics; they establish no exhaustive clause coverage or required implementation technology.

Agent controlRelevant topicEvidence to prepare
Authority and pre-action policyprEN 18228: risk managementRisk scenario; delegated scope; policy version; denied-action test
Tool and data scopeprEN 18282: cybersecurityThreat model; credential and network boundaries; unauthorized-access tests
ValidationprEN 18228 / prEN 18282Test set; known failure modes; results; reviewer and residual-risk decision
Human escalationprEN 18228; assess applicable human-oversight requirements separatelyReviewer authority; context shown; rejection, expiry and escalation tests
MonitoringprEN 18228 / prEN 18282Control-effectiveness review; incident triage; remediation and re-test
EvidenceprEN 18229-1: loggingEvent schema; decision/execution linkage; access and retention rules; reconstruction test
EN 18286

Place the register inside quality management

Assign an owner for the register, approve changes, retain review results and define revalidation triggers. Record the intended use, system boundary and applicable provider or deployer role before assessing obligations.

For each control, add the exact edition and clause when an authorized reviewer has assessed the standard text. Record gaps and compensating measures explicitly. Keep procurement statements consistent with the completed assessment.

Logging example

Build an investigation record

For an illustrative AML alert-closure request, retain the tenant and agent identity, case identifier, authorized action, policy version and decision, approval identity where required, execution result, timestamps and correlation identifiers. Define redaction and access rules so the record does not unnecessarily replicate sensitive case data.

Test whether a reviewer can follow a rejected request and an executed request from start to finish. Check missing records, retries and partial failures. Cryptographic sealing is a separate implementation choice to assess; this guide makes no claim that the logging draft mandates it.

Conformity

Keep the legal claim bounded

An Official Journal reference and conformity with the provisions covering the relevant legal requirements determine the presumption available. Publication, a product mapping or a completed demonstration alone cannot establish it.

KLA Control Plane can contribute policy and approval records, execution lineage and evidence artifacts from integrated paths. Organizational governance, system classification, validation and the applicable conformity assessment remain part of the wider work.

Use the architecture guide to assign implementation boundaries before turning this matrix into a procurement checklist.

FAQ

Questions buyers should resolve

Is this a clause-by-clause standards crosswalk?

This is a topic-level preparation map using public standards metadata. It contains no claim to reproduce or fully assess the draft normative requirements.

Does mapping KLA controls establish conformity?

A mapping identifies candidate controls and evidence. Conformity requires assessment of the system against applicable requirements and the relevant standard provisions, with any presumption limited by the Official Journal reference.

Links

Related links

Regulated Agent Harness Architecture

/guides/regulated-agent-harness-architecture

Open

Bank of England harness engineering analysis

/blog/bank-of-england-ai-harness-engineering

Open

Preparing agent controls for EU AI Act standards

/blog/eu-ai-act-standards-agent-controls

Open

SAFR runtime framework

/blog/safr-mas-framework-explained

Open

Discuss your agent control architecture

/book-demo

Open
AI Act standards: agent control mapping | KLA