1. The Statutory Text
Article 4 of Regulation (EU) 2024/1689 (the EU AI Act) entered into force on 2 February 2025. There was no grace period. The obligation reads:
Providers and deployers of AI systems shall take measures to ensure, to their best extent, a sufficient level of AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf, taking into account their technical knowledge, experience, education and training and the context the AI systems are to be used in, as well as the persons or groups of persons on whom the AI systems are to be used.
— EU AI Act, Article 4 · Regulation (EU) 2024/1689 · OJ L 2024/1689Article 4 is not a forthcoming obligation. It is in force now. Any organisation that has deployed AI systems and has not documented its staff AI literacy position is already in breach, even where those systems are general-purpose tools (email, drafting assistants, scheduling AI) rather than Annex III high-risk systems.
2. Three Operative Elements
Element 1 — "Providers and deployers"
The obligation falls on both the entity that provides the AI system and the entity that deploys it. For regulated sectors, the deployer is almost always the regulated firm itself — the law firm, the financial institution, the healthcare trust. The deployer cannot delegate compliance to its AI vendor.
UK application: The EU AI Act does not apply directly in the UK post-Brexit. However, SRA-regulated firms serving EU clients or operating under EU-connected mandates are subject to extraterritorial provisions. Moreover, the SRA's supervisory scope expressly covers AI governance and staff competence. Article 4 is the evidential standard that UK regulators will reference when framing their own supervisory questions.
Element 2 — "Sufficient level… shall be documented"
The phrase "to their best extent" does not grant discretion over whether to comply — it sets the effort standard. "Sufficient level" is relative to: the technical complexity of the AI system in use; the role of the staff member operating it; the context in which the system is deployed; and the population the system affects. What is "sufficient" for a partner reviewing a generative AI legal research tool differs from what is sufficient for a data entry clerk whose workflow is partially automated.
Crucially, sufficiency must be demonstrated. A policy stating that training "will be provided" does not constitute evidence of compliance. A log recording what training was provided, to whom, when, and at what assessed competency level does.
Element 3 — "Persons dealing with the operation and use of AI systems on their behalf"
The obligation extends to contractors, secondees, and third parties who operate the firm's AI systems. A firm that uses a managed AI service delivered by a contractor workforce must ensure that those individuals also meet the literacy threshold. Panel client governance frameworks increasingly replicate this requirement, exposing firms to dual regulatory and commercial risk.
Scenario: Halstead & Cole LLP deploys AIMS-SYS-003 (a legal research assistant) and AIMS-SYS-004 (a document drafting accelerator). COLP Priya Anand is required to confirm, when Pemberton Capital LLP issues its AI governance questionnaire, that all fee earners using those systems have completed and evidenced AI literacy training. Without REG-AIMS-LIT-001, no such confirmation can be made with regulatory integrity.
3. The Three-Phase Enforcement Timeline
| Date | Provision | Obligation | Status |
|---|---|---|---|
| 2 Feb 2025 | Article 4 — AI Literacy | All deployers shall ensure and document sufficient AI literacy for staff operating AI systems. No grace period applies. | ● ACTIVE NOW |
| 2 Dec 2026 | Article 50 — Transparency | Certain AI systems must disclose their AI nature to users. Includes GPAI model obligations on providers. | ● 15 months |
| 2 Dec 2027 | Annex III — High-Risk (standalone systems) | High-risk AI systems listed in Annex III face full conformity obligations including fundamental rights impact assessment. | ● 27 months |
| 2 Aug 2028 | Annex III — High-Risk (embedded systems) | High-risk AI embedded in regulated products face the same obligations; extended timeline to allow product update cycles. | ● 35 months |
Article 4 is a gateway obligation. Regulators auditing an organisation's Annex III high-risk compliance will treat the absence of Article 4 literacy records as evidence that the broader compliance programme lacks adequate foundations. Failure to establish REG-AIMS-LIT-001 now undermines every subsequent compliance assertion made in 2027 and 2028.
4. ISO/IEC 42001:2023 Clause 7.2 — Competence
ISO/IEC 42001:2023, the international standard for AI management systems, requires at Clause 7.2 that organisations:
- a)Determine the necessary competence of persons doing work under its control that affects its AI management system performance;
- b)Ensure those persons are competent on the basis of appropriate education, training, or experience;
- c)Where applicable, take actions to acquire the necessary competence and evaluate the effectiveness of those actions;
- d)Retain appropriate documented information as evidence of competence.
ISO/IEC 42001 provides the framework. REG-AIMS-LIT-001 is the evidence instrument that converts framework into regulatory proof. Clause 7.2(d) is satisfied by a populated, COLP-verified training register. It is not satisfied by a training policy alone.
5. ITIL 4 Service Management Mapping
Under ITIL 4's Workforce and Talent Management Practice, Article 4 compliance is correctly categorised as a recurring service demand event rather than a one-time project. This has three operational implications:
Each new AI system deployment creates a training demand event. The service catalogue entry for AI literacy training is triggered by system onboarding, not by regulatory deadline.
Literacy is not a one-time certification. A 12-month refresh cadence is the minimum viable standard for fast-moving AI system contexts. WF-LIT-003 automates overdue detection.
Where evidence exists (vendor attendance records, email confirmations, policy acknowledgements), retroactive logging is permissible. A firm that has been training staff informally may have more to log than it realises.
6. The Evidence Gap
The policy tells regulators what the firm intends. The training record tells them what the firm did. When a regulator asks for evidence under Article 4 or a panel client issues a Section 7 AI governance questionnaire, only the second document answers the question. A firm with a well-drafted AI literacy policy and no training log has demonstrated intent, not compliance.
The five evidence fields that REG-AIMS-LIT-001 must capture to satisfy an Article 4 inspection are:
Part B of this pack provides the complete training log template with all REG-AIMS-LIT-001 fields, field-by-field completion guidance, and the competency assessment matrix. Part C provides the COLP verification checklist and sign-off protocol.
Training Record — REG-AIMS-LIT-001
Competency Level Reference Matrix
Training records shall be retained for a minimum of seven (7) years from the completed_date, in accordance with UK Companies Act 2006 record retention obligations and aligned to the EU AI Act's supervisory inspection windows. Records may be destroyed thereafter unless a regulatory investigation is ongoing or the individual is the subject of a professional disciplinary proceeding.
Only the designated COLP, DPO, or an officer formally delegated in the firm's AI governance policy may complete this checklist. Peer verification (e.g., the training record subject verifying their own record) is not permissible and will be flagged as a compliance gap by WF-LIT-003. Delegation must be documented and the delegate's staff code must appear in colp_verified_by.
Verification Record Header
Five-Point Verification Checklist
Review each point against the submitted training record. Mark each point as PASS, FAIL, or QUERY (requires further evidence before a determination can be made). A single FAIL result means the record cannot be verified and must be returned for correction.
Verification Outcome and Sign-Off
All five verification points have been assessed as PASS. The verifier confirms the record is accurate, the evidence is adequate, and the competency level is appropriate. The record shall be marked colp_verified = TRUE in REG-AIMS-LIT-001.
One or more verification points have been assessed as FAIL or remain QUERY after 5 working days. Record is returned to the submitter with the reason below. A corrected record must be re-submitted and re-verified before it contributes to compliance status.
Where the firm operates the AIMS automation layer, the verification outcome shall also be submitted via the WF-LIT-002 COLP Verification Gate webhook (POST /webhook/lit/verify-record) using the x-aims-colp-key authentication header. The manual sign-off above serves as the primary audit trail; the digital submission updates REG-AIMS-LIT-001 automatically and triggers confirmation emails to the submitter and COLP.
A staff member using an AI system while their training record is in "NOT LOGGED", "Overdue", or "Refresh Required" status creates an active Article 4 breach for every day that gap persists. The monitoring procedure below is not a best-practice recommendation — it is the minimum viable process for maintaining compliance once REG-AIMS-LIT-001 is operational.
1. Monitoring Schedule
2. The Three Compliance Gap Categories
The monitoring procedure tracks three distinct gap states, each requiring a different response. These categories correspond to the status values in the vw_lit_compliance_gaps database view.
3. Daily Monitoring Procedure — Step by Step
The daily schedule trigger fires at 08:00. The sp_flag_overdue_training() stored procedure is called against the REG-AIMS-LIT-001 database. This procedure updates the record_status field for all active records where: (a) completed_date exists but refresh_due_date has passed → status set to "Overdue"; (b) no training record exists for a registered AI system user → gap flagged as "NOT LOGGED"; (c) refresh_due_date is within 30 days → status set to "Refresh Required".
The vw_lit_compliance_gaps view is queried to retrieve all records with status: NOT LOGGED, Overdue, or Refresh Required. Results are grouped by gap status and sorted by severity (NOT LOGGED first, Overdue second, Refresh Required third). If the query returns zero rows, no digest is sent and the workflow terminates cleanly (no-gap confirmation is logged internally).
Where gaps exist, the workflow constructs a structured digest. The digest header contains: date, total gap count, breakdown by status category, and the firm_id. Each gap record includes: staff_code, staff_name, role_title, ai_system_ref, gap_status, days_since_due (for Overdue), days_until_due (for Refresh Required). For NOT LOGGED gaps, the date the staff member was registered as a system user is included.
The digest is delivered to the COLP's designated compliance inbox. The email subject line includes the gap count and highest-severity category to enable triage without opening: e.g., "REG-AIMS-LIT-001 Daily Digest — 3 gaps (2 NOT LOGGED, 1 Refresh Required) — 01 Sep 2026". The email is sent only when total_gaps > 0; the workflow is silent on zero-gap days.
Upon receiving the digest, the COLP shall: review each gap entry; for NOT LOGGED items — contact the relevant line manager and L&D team to schedule training within 5 working days and consider whether the staff member's AI system access should be restricted pending training completion; for Overdue items — initiate the refresh booking and update the risk register; for Refresh Required items — confirm that a booking is in the L&D pipeline. The COLP shall record the action taken against each gap in the risk register within the same business day.
The COLP notifies the relevant L&D lead and line manager of each gap. Notification shall include: the staff member's name and code, the AI system reference, the gap status, and the required action deadline. For firms using the AIMS automation layer, a supplementary WF-LIT-001 pre-populated draft can be sent to the future training provider to accelerate logging on completion.
A gap is closed when: (a) a new training record is submitted via WF-LIT-001 and the completed_date populates; (b) the COLP verifies the record via WF-LIT-002; (c) the record_status updates to "Verified". The gap will no longer appear in the vw_lit_compliance_gaps view for that staff/system combination. The audit log (lit_audit_log) retains a full history of gap detection, escalation, and closure events for regulatory inspection purposes.
4. Escalation Thresholds
5. Record Keeping for This Procedure
The COLP shall maintain a Monitoring Action Log documenting: the date of each daily digest received, the gap count by category, and the action(s) taken. This log is a separate document from REG-AIMS-LIT-001 and serves as evidence that the monitoring procedure was operated in good faith. In a regulatory inspection, the Monitoring Action Log demonstrates proactive management, not merely passive record keeping.
Where a submission is made to a regulatory body as part of an enforcement investigation (as opposed to a routine audit or self-assessment), firms should obtain legal advice before submitting. This cover sheet is a presentation aid; it does not constitute legal advice and does not determine what must or should be disclosed.
Evidence Pack Cover Sheet
Section 1 — Submitting Organisation
Section 2 — Submission Details
Section 3 — Document Index
Section 4 — Compliance Status Declaration
The undersigned, being the [COLP / DPO / Chief Compliance Officer] of [ORGANISATION NAME], hereby confirms that:
- The organisation has taken measures to ensure a sufficient level of AI literacy among staff and persons operating AI systems on its behalf, in accordance with Article 4 of Regulation (EU) 2024/1689.
- The training records presented in this evidence pack are accurate and complete to the best of the signatory's knowledge as at the date of submission.
- All records marked "COLP Verified" in REG-AIMS-LIT-001 have been reviewed by an authorised officer and the supporting evidence was available and adequate at the time of verification.
- Open compliance gaps identified by the monitoring procedure (if any) are identified and described in the accompanying gap schedule, together with the remediation actions taken or planned.
- The organisation understands that the submission of inaccurate, misleading, or incomplete information to a regulatory authority may constitute a regulatory breach independently of the underlying compliance position.
Guidance: Presenting the Evidence Pack to a Panel Client
Panel clients issuing AI governance questionnaires (such as the Pemberton Capital LLP Section 7 questionnaire scenario) are not regulators and the submission does not carry formal legal privilege considerations. However, the same discipline applies: the training register extract and verification checklists should be presented as evidence of what was done, not as assertions of what the firm intends.
When the questionnaire asks: "What AI literacy training have staff completed?"
Provide the REG-AIMS-LIT-001 extract filtered to the systems relevant to the client's matter. Do not provide a policy document in response to a training completion question — it answers a different question.
When the questionnaire asks: "How does your firm ensure AI literacy obligations are met on an ongoing basis?"
Describe the three-workflow monitoring cycle (WF-LIT-001 intake → WF-LIT-002 verification → WF-LIT-003 daily monitoring) and provide the Monitoring Procedure (Part D of this pack) as a supporting document.
Where the organisation is pursuing ISO/IEC 42001:2023 certification, this evidence pack satisfies the documented evidence requirement at Clause 7.2(d). Provide it to the certification auditor as tab D of the Clause 7.2 evidence bundle, together with the AI Management System competency needs assessment (if separately prepared).