Independent thinking. Informed defence.

CISO Times

Intelligence for the people behind the defence.

The CISO Decision Brief

When do we…

When must staff who use AI have training, and what must it cover?

9 decision pointsThe pack behind this answer: AI Security Governance & Acceptable Use

Add to board memo

The short answer

Recommendation
Train everyone who uses, oversees or connects AI for you before their access is switched on, with one short essentials module, and add a module for each role that needs more. Keep an internal record of what each person completed.
Reasoning
Article 4 of the EU AI Act, as replaced by Regulation (EU) 2026/1744, asks providers and deployers to take measures that support the AI literacy of their staff and others who operate or use AI systems for them. The measures must fit each person's knowledge and experience, the context and the people the system is used on. It asks for measures. It asks for no certificate and no guaranteed level for any person. So the plan gives every user the essentials and adds modules for owners, customer-facing staff, publishers, HR, builders and the management body.
Considerations
  • The timings are the pack's defaults, and Article 4 sets no figure: essentials within 60 calendar days of launch for people who already have access, refreshed every 12 months, and a new member of the management body briefed within 10 working days (AG-18).
  • Article 4 has applied since 2 February 2025, and national market surveillance authorities supervise it from 2 August 2026, as the pack states. Counsel confirms whether it reaches you as a provider or a deployer.
  • The AI plan sits beside your baseline security awareness course and does not replace it.
  • Report counts with the gap named, never a percentage or a score. Group results go only to groups large enough that no one can be singled out.

The decision, step by step

Select any step for the reasoning behind it and where the pack covers it.

  • Start
  • Step
  • Decision
  • Outcome
StartStaff are using, or about to use, AI for workDecisionDo they use, oversee or connect AI for us?OutcomeNo measure yet: recheck when access or role changesDecisionDid they already have access when the plan started?StepFinish AI essentials within the plan's days of launch (60 by default)StepFinish AI essentials before the account is switched onDecisionDo they own an AI use or check its output?StepAdd owning and overseeing an AI use (AL-02)DecisionDo they answer customers or hand over from an AI assistant?StepAdd customers and AI (AL-04) before go-liveDecisionDo they publish content made with generative tools?StepAdd publishing AI-generated content (AL-05)DecisionDo they recruit, appraise or discipline, or use AI on people?StepAdd AI and decisions about people (AL-07)DecisionDo they write code with AI, connect plug-ins or set up an AI system?StepAdd building with AI safely (AL-03)DecisionAre they on the management body, or do they take Tier 3 decisions?StepAdd the management briefing (AL-06)StepKeep an internal record of what each person completedDecisionIs anyone neither done nor within their time?OutcomeFix it with the audience owner, set a date, report the gapOutcomePlan, record and counts kept; review when tools or law change

Keyboard: Tab to a step, arrow keys move along the flow, Enter pins its explanation, + and − zoom, 0 fits the diagram, Esc closes. The text version below has the same content.

Where a step quotes a number, it is the default in the CISO Times pack. Your own standard may set a different one.

Read as text

The same flow written out: each step, its reasoning and where the pack covers it. 22 steps.

  1. Start

    Staff are using, or about to use, AI for work

    Article 4, as replaced by Regulation (EU) 2026/1744, asks providers and deployers to take measures to support the AI literacy of their staff and other persons dealing with the operation and use of AI systems on their behalf. The measures take account of knowledge, experience, education and training, the context of use and the people the systems are used on. It does not require a guaranteed level for any individual. Article 4 has applied since 2 February 2025, and the replaced wording since 27 July 2026.

    In the pack

    See also

  2. Decision

    Do they use, oversee or connect AI for us?

    The duty covers staff and other persons who deal with the operation and use of AI systems on the organisation's behalf. In the pack that includes contractors and supplier staff who hold your accounts (AU-1). Judgement: someone with no access to an AI tool and no part in any use is outside the plan until that changes. If not known, take the Yes branch.

    In the pack

  3. Outcome

    No measure yet: recheck when access or role changes

    Revisit when their access, role or tools change. A new tool goes through intake first, and the plan is reviewed whenever a new tool, a new audience or a change in the law changes who needs what.

    In the pack

    See also

  4. Decision

    Did they already have access when the plan started?

    The Article 4 text in the pack sets no deadline for each person, so the timing is the pack's default. People who already have access finish within a set number of calendar days from the plan's launch (AG-18). Anyone else finishes before the account is switched on. If not known, check the access list and take the second branch, which is the stricter.

    In the pack

  5. Step

    Finish AI essentials within the plan's days of launch (60 by default)

    AL-01 is the AI essentials module: what an AI system is and is not, how generative tools can be wrong, the approved list and data limits, signing in with the work account, checking output, what to disclose and what to report. Existing users finish within the days set from launch (60 by the pack's default in AG-18) and repeat it every 12 months. It sits beside the baseline security awareness course and does not replace it.

    In the pack

  6. Step

    Finish AI essentials before the account is switched on

    A joiner with access, or a new user of an approved tool, completes AL-01 first (AG-18). Intake switches access on for people who have completed it (IS-6). They repeat it every 12 months by the pack's default, the cadence of the baseline security course.

    In the pack

  7. Decision

    Do they own an AI use or check its output?

    Every named owner and overseer in the inventory is in this audience (AU-2). Owners answer for the use and its output, and overseers must be competent to judge the output and have authority to reject it. If not known, take the Yes branch.

    In the pack

  8. Step

    Add owning and overseeing an AI use (AL-02)

    Owners and overseers take AL-02 before the first approval they hold, then every 12 months. It covers the intake questions, the tiers and the inventory, what human oversight means and how to do it, reviews and triggers, and what a competent overseer checks. For high-risk systems the Act asks for oversight by people with the necessary competence, training and authority (Article 26(2), from 2 December 2027 for Annex III).

    In the pack

  9. Decision

    Do they answer customers or hand over from an AI assistant?

    Customer service staff who would take over from, or correct, a customer-facing assistant are this audience (AU-3). If no customer-facing assistant exists or is planned, the answer is No. If not known, take the Yes branch.

    In the pack

  10. Step

    Add customers and AI (AL-04) before go-live

    AL-04 covers telling customers they deal with an AI system, how to take over from an assistant, what never to leave to it and how to record a complaint about its answer. It is due before any customer-facing assistant goes live, then every 12 months.

    In the pack

  11. Decision

    Do they publish content made with generative tools?

    Marketing, sales and anyone who publishes text, images, audio or video outside the organisation are this audience (AU-4). If not known, take the Yes branch.

    In the pack

  12. Step

    Add publishing AI-generated content (AL-05)

    AL-05 is due before they use a generative tool for published work. It covers labelling realistic AI images, audio, video and text, using no real people without consent, and checking facts, rights and tone. The labelling link is Article 50(4), which applies from 2 August 2026.

    In the pack

  13. Decision

    Do they recruit, appraise or discipline, or use AI on people?

    HR staff, and managers who recruit, appraise or discipline, are this audience (AU-6). A person decides any matter that affects someone's job, pay, discipline or legal position, and a use that decides or shapes such a matter is Tier 3 (AG-06). If not known, take the Yes branch.

    In the pack

  14. Step

    Add AI and decisions about people (AL-07)

    AL-07 is due before any use that touches a decision about a person. It covers why a person decides (GDPR Article 22), the Annex III employment area, the prohibition on inferring emotions at work, telling and consulting staff, and the rights of candidates and staff.

    In the pack

  15. Decision

    Do they write code with AI, connect plug-ins or set up an AI system?

    Developers and IT administrators who write code with an AI assistant, connect plug-ins or set up an AI system are this audience (AU-7). If not known, take the Yes branch.

    In the pack

  16. Step

    Add building with AI safely (AL-03)

    AL-03 is due before they connect or build. It covers keeping secrets out of prompts, reviewing AI-generated code, plug-ins and agents that act, the supplier terms, and prompt injection.

    In the pack

  17. Decision

    Are they on the management body, or do they take Tier 3 decisions?

    Members of the management body and whoever your procedure names to take Tier 3 decisions are this audience (AU-5): executive management by default, or the board or its equivalent where your governance structure puts it. Who sits at the top depends on how you are governed: the board or a delegated committee, the management board in a two-tier structure, or executive management where there is no board. If not known, take the Yes branch.

    In the pack

  18. Step

    Add the management briefing (AL-06)

    AL-06 is a briefing when the policy is approved and every 12 months by the pack's default. A new member of the management body takes it within 10 working days of joining. It covers the pack's idea, the tiers, the Tier 3 decision, what the Act asks of a deployer and when, the prohibitions, and the measures the management body sees.

    In the pack

  19. Step

    Keep an internal record of what each person completed

    The Commission says no certificate is needed and an internal record of trainings or other guiding initiatives is enough. Record the person, audience, module, module version, due date, completion date, how delivered, evidence reference and next due date. Treat the record as personal data. Keep it with the plan for 3 years after the use ends, or longer where law or a contract asks (the pack's default).

    In the pack

  20. Decision

    Is anyone neither done nor within their time?

    Count each audience as complete, not yet due or overdue. A person in two audiences is one entry in each, so the entry totals are not a count of people. Report counts with the gap named, never a percentage or a score (AG-22). If not known, take the Yes branch.

    In the pack

  21. Outcome

    Fix it with the audience owner, set a date, report the gap

    Close an overdue audience with its owner and a date, and report the gap with the counts. Group results go only to groups of at least 10 people (P16 AW-13), so a small audience is reported with another.

    In the pack

  22. Outcome

    Plan, record and counts kept; review when tools or law change

    Review the plan whenever a new tool, a new audience or a change in the law changes who needs what, and refresh AL-01 every 12 months by the default. National market surveillance authorities supervise Article 4 from 2 August 2026. Judgement: keep the plan, module versions and record where you can show them.

    In the pack

    See also

Practical checklist

  1. List everyone who uses, oversees or connects AI for you, including contractors and supplier staff who hold your accounts.
  2. Have new users finish AI essentials before access is switched on, and existing users within the plan's days of launch.
  3. Add the owner and overseer module for anyone who owns an AI use or checks its output.
  4. Add the role modules before the activity starts: customer handover, publishing generated content, decisions about people, and building or connecting AI.
  5. Brief the management body and whoever takes Tier 3 decisions when the policy is approved and each year.
  6. Record each completion with person, audience, module, version, due date, completion date, delivery, evidence and next due.
  7. Count each audience as complete, not yet due or overdue, and close each overdue audience with its owner and a date.

Evidence to have ready

  • The literacy plan with its audiences, modules and dates, approved and under document control.
  • The version of each module in use and the date it last changed.
  • The completion record, one row per person and module, kept as personal data.
  • The management briefing's date, attendees and slides or notes.
  • The month-end counts by audience, and the list of people whose access was switched on after completion.

Assumptions

  • Article 4 reaches you as a provider or a deployer. Most organisations that use other people's AI are asking the deployer question, and counsel confirms it.
  • The timings are the CISO Times literacy plan's defaults: 60 calendar days, 12 months and 10 working days for a new member of the management body. Article 4 sets no figure, and your own standard may differ.
  • The AI plan sits beside your baseline security awareness course and does not replace it.
  • National market surveillance authorities supervise Article 4 from 2 August 2026, and your Member State's rules may add to it.

Not covered here