Independent thinking. Informed defence.

CISO Times

Intelligence for the people behind the defence.

The CISO Decision Brief

Who decides…

Who is accountable for an AI decision, and who can stop an AI use?

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

Add to board memo

The short answer

Recommendation
Give each AI activity exactly one accountable role, and let nobody approve a use they own. Tier approvers answer for whether a use starts, the AI use owner answers for its output, and the AI governance lead may suspend an approved running use pending review.
Reasoning
A matrix with one accountable role per activity stops a decision falling between roles (RS-01 to RS-16). The tier fixes who may approve: the AI governance lead for Tier 1; the lead with the Head of Information Security for Tier 2; the executive decision-maker your procedure names for Tier 3. Where an approver owns the use, the executive decision-maker takes that approver's place (RS-16). No AI output decides a matter about a person alone: a named person decides after reading the evidence (AG-06). Work can be delegated, accountability cannot.
Considerations
  • Only the board, or a board risk committee it has delegated to, may accept a Critical factor or one outside tolerance; the management board in a two-tier structure; executive management where there is no board.
  • If the executive decision-maker owns a Tier 3 use, the board or its equivalent decides it. In a small organisation, ask an outside adviser or a non-executive director to read the file.
  • Where a service provider runs your IT, it can do the work. The accountability stays inside your organisation.
  • A prohibited practice or special category data is stopped now. Other unapproved uses are stopped within 3 working days by the pack's default (AG-02).

The decision, step by step

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

  • Start
  • Step
  • Decision
  • Outcome
StartSomeone asks who answers for an AI decision, or who can stop a useDecisionWhat is the question about?DecisionWhich tier did the screen set?OutcomeTier 1: the AI governance lead answersOutcomeTier 2: the AI governance lead and the Head of InfoSec answerDecisionDoes the assessment leave a Critical factor, or one outside tolerance?OutcomeThe board or its equivalent must accept the riskOutcomeTier 3: the executive decision-maker answersDecisionDoes the use decide, shape or inform a decision about a person?OutcomeA named person decides; the use is Tier 3, or Tier 2 if it informsOutcomeThe AI use owner answers for the outputDecisionWhy does the use need stopping?DecisionIs it a prohibited practice, or special category data going in?OutcomeStop it now; the AI governance lead tells Legal or the DPO todayOutcomeStop it within the policy's working days and help people moveOutcomeThe approval switches it off, and the approvers are toldOutcomeReport it through the incident route and contain itOutcomeThe AI governance lead suspends it; the tier's approvers decide

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. 18 steps.

  1. Start

    Someone asks who answers for an AI decision, or who can stop a use

    Accountability for AI is set in a matrix: each of sixteen activities has exactly one accountable role (A), plus roles that do the work (R), are consulted (C) and are told (I). A role holder may delegate the work to a named person, never the accountability. Find the activity you are asking about.

    In the pack

  2. Decision

    What is the question about?

    Three situations cover most questions. A proposed use needs a decision on whether it starts. A decision about a person or an output needs someone to answer for it. A use that is already running may need stopping. Pick the nearest and follow it.

    In the pack

  3. Decision

    Which tier did the screen set?

    The tier fixes who may approve. Only the approvers named for the tier may approve, and any of them may hold or decline. If the tier is not yet set, the AI governance lead sets it first. Take the higher tier if you cannot tell.

    In the pack

    See also

  4. Outcome

    Tier 1: the AI governance lead answers

    The AI governance lead screens and approves Tier 1 uses and answers for the decision (RS-05). The Head of Information Security is consulted. Nobody approves a use they own: if the lead owns the use, the executive decision-maker approves in the lead's place (RS-16). In a small organisation one person often holds several roles, so keep the person who proposes a use apart from the person who approves it.

    In the pack

  5. Outcome

    Tier 2: the AI governance lead and the Head of InfoSec answer

    The AI governance lead is accountable for a Tier 2 approval, with the Head of Information Security (RS-06). The AI use owner is consulted and signs the conditions it accepts, but does not approve: nobody approves a use they own (IS-4). Where an approver owns the use, the executive decision-maker takes that approver's place (RS-16). The Data Protection Officer's advice is recorded where personal data is used.

    In the pack

  6. Decision

    Does the assessment leave a Critical factor, or one outside tolerance?

    The executive decision-maker decides Tier 3 uses, and signs as well for a High risk outside appetite. A Critical factor, or one outside tolerance, goes further: only the board or its equivalent may accept it (RM-04). If the assessment is unfinished, take the first branch.

    In the pack

    See also

  7. Outcome

    The board or its equivalent must accept the risk

    The acceptor is the board, or a board risk committee it has delegated this to; in a two-tier structure, the management board; where there is no board, such as in an owner-managed company, executive management. Where a board exists, executive management can recommend, reduce the risk or decline, but it cannot accept a Critical factor (RM-04). The intake procedure's executive role does not override this.

    In the pack

    See also

  8. Outcome

    Tier 3: the executive decision-maker answers

    The executive decision-maker the procedure names is accountable for approving or declining a Tier 3 use (RS-07): executive management by default, or the board where your governance structure puts it. It acts on the AI governance lead's recommendation. If the decision-maker owns the use, the board or its equivalent decides (D-42). Legal, HR and the Data Protection Officer are consulted. In a small organisation the managing director is often both lead and decision-maker, so ask an outside adviser or a non-executive director to read the file.

    In the pack

  9. Decision

    Does the use decide, shape or inform a decision about a person?

    Examples are recruitment, pay, performance, discipline, credit and access to a service. No AI output decides such a matter alone (GDPR Article 22). A named person decides after reading the evidence. A use that decides or shapes the decision is Tier 3; a use that only informs it is Tier 2; a use in an Annex III area is Tier 3 whatever it does (AG-06). If you are not sure, take the first branch.

    In the pack

  10. Outcome

    A named person decides; the use is Tier 3, or Tier 2 if it informs

    The decision is made by a named person who has read the evidence and has the authority to reject the output. A use that decides or shapes the decision is Tier 3 and is not started before the Tier 3 decision, which the executive decision-maker answers for (RS-07). A use that only informs it is Tier 2 (RS-06). HR advises, and Legal confirms the reading of the EU AI Act. The file records the name of the person who makes the decision.

    In the pack

    See also

  11. Outcome

    The AI use owner answers for the output

    The AI use owner is accountable for checking the output and answering for it (RS-12). The user does the checking: a person competent to judge the output reads it before it is relied on, sent or published, and may reject it. The owner is the manager accountable for the business activity, not the person who switched the tool on.

    In the pack

  12. Decision

    Why does the use need stopping?

    Three routes stop a use that is not approved or has broken its approval: the unapproved-use steps, a dated condition and the incident route. For an approved use running without one of these causes, the AI governance lead may suspend it pending review (RS-15). If the reason is none of these, take the last branch.

    In the pack

  13. Decision

    Is it a prohibited practice, or special category data going in?

    Both are a Stop. They are stopped now, without waiting for the working days. The AI governance lead protects first and tells Legal for a prohibited practice or the Data Protection Officer for special category data the same day. Take the first branch if you do not know.

    In the pack

  14. Outcome

    Stop it now; the AI governance lead tells Legal or the DPO today

    No dated condition is allowed. For special category data, report at once through the incident route and tell the Data Protection Officer the same day. Judgement: the Head of IT applies the technical switch-off, since the matrix makes the Head of IT accountable for the tool's settings and single sign-on (RS-10).

    In the pack

    See also

  15. Outcome

    Stop it within the policy's working days and help people move

    An unapproved use found is stopped within 3 working days of discovery by default (AG-02). The AI governance lead and the AI use owner act, and the people involved are helped to an approved tool. If Internal or Confidential data went into the tool, report it as an event through the incident route. Nobody is blamed for a report made in good faith.

    In the pack

    See also

  16. Outcome

    The approval switches it off, and the approvers are told

    A dated condition is part of the approval, and the tool is switched off if it is missed. A condition not met is reported to the approvers. Judgement: the Head of IT applies the technical switch-off (RS-10), and the AI governance lead records the new status in the inventory. Moving the date is a departure from the approval and goes through the exception process (AG-20), not a quiet extension.

    In the pack

  17. Outcome

    Report it through the incident route and contain it

    Report an AI-related event at once through the incident route, which decides whether it is an incident and holds the clocks (AG-13). For a manipulated input that made a system leak data or act wrongly, contain it first where it can still leak or act, and capture what you can in the same step; otherwise preserve the evidence first and then switch off its access (AT-6, P09 IM-07). The Head of Information Security is accountable for the hand-over (RS-13). The incident pack, not this flow, decides what follows.

    In the pack

  18. Outcome

    The AI governance lead suspends it; the tier's approvers decide

    The AI governance lead may suspend an approved running use pending review, and answers for doing so (RS-15, AG-10). The Head of IT applies the switch-off, and the use owner is consulted. The approvers of the use's tier decide whether it resumes, retires or continues on new conditions. A trigger forces the review within 10 working days by default.

    In the pack

Practical checklist

  1. Find the activity in the responsibility matrix and its one accountable role.
  2. For a proposed use, check the tier the screen set and its named approvers.
  3. Send a Critical factor, or one outside tolerance, to the board or its equivalent.
  4. For a decision about a person, name the person who decides after reading the evidence.
  5. Stop an unapproved use now if it is a prohibited practice or takes special category data, and tell Legal or the Data Protection Officer the same day.
  6. Switch off a use whose dated condition was missed, and tell its approvers.
  7. Report an AI-related event at once through the incident route.
  8. Suspend an approved running use pending review for any other reason, and let the tier's approvers decide whether it resumes.

Evidence to have ready

  • The responsibility matrix, with one accountable role on each activity and the role holders named.
  • Each decision record, showing the tier's approvers by name and that the use owner is not among them.
  • The name of the person who decides each matter about a person that an AI use decides, shapes or informs.
  • The record of any suspension, stop or switch-off, and the approvers' decision on whether the use resumes.

Assumptions

  • You have named, or will name, the roles: an AI governance lead, an AI use owner for each use, a Head of Information Security and an executive decision-maker. Where one person holds several roles, keep the single accountable role on each activity.
  • The approvers depend on how you are governed. The top approver is the board, or a board risk committee it has delegated to; the management board in a two-tier structure; executive management where there is no board. Your procedure names who takes Tier 3 decisions: executive management by default, or the board.
  • Where a service provider runs your IT, it can do the work but the accountability stays inside your organisation.
  • The tiers and the working days to stop an unapproved use are the pack's defaults. Your own standard may set different ones.

Not covered here