If staff are already using AI tools nobody approved, then what do we do?

Start

We find staff using AI tools nobody approved

Finding one means the procedure is working. Most organisations find AI uses they did not know about: a free assistant on personal accounts, a feature a supplier added in a release, or a tool a team bought with a card. The aim is to make the use safe first and then decide whether it can go on. Do not email everyone to ask who uses AI and threaten a penalty, because people hide what they do.

• Then: Tell the AI governance lead the same day

Step

Tell the AI governance lead the same day

Whoever finds the use tells the lead, whatever the source: the inventory survey, discovery of systems bought with a card or signed into with a work address, a supplier's notice of a new AI feature, a manager's question or a staff report. In the pack's example the staff survey found the free assistant on personal accounts. The lead owns the steps that follow.

• Then: Never-allowed practice or special category data, as far as you know?

Decision

Never-allowed practice or special category data, as far as you know?

These two cases skip the working days the policy otherwise allows. A use that matches a prohibited practice, or that takes special category personal data, stops now. Tell Legal the same day for a prohibited practice. For special category data report at once and tell the Data Protection Officer the same day. Legal decides whether a use matches and whether anything further is needed. Judgement: if you are unsure, answer yes. The questions that follow check again.

• Yes, or not sure: Stop it now and tell Legal or the DPO the same day
• No: Ask people to put nothing above Public, and no personal data, in

Outcome

Stop it now and tell Legal or the DPO the same day

Do not wait for the working days the policy sets, and set no dated condition. Tell Legal the same day for a prohibited practice. For special category data, report it at once as an event through the incident route and tell the Data Protection Officer the same day. Nobody is blamed for a report made in good faith, and the incident pack decides whether it is an incident and what must be reported.

Step

Ask people to put nothing above Public, and no personal data, in

Do this first, before you work out the rest. It limits the exposure while you find the facts. Say plainly that nobody is blamed for a report made in good faith, because people who feel blamed hide the use. Public information with no personal data in it stays allowed in any tool, but never through a work account. Personal data needs a tool approved for it, even inside Public information (BC-5).

• Then: Ask what went in, over what period, through which account

Step

Ask what went in, over what period, through which account

Compare the answers with the approved list. The questions are about data level, not fault. Ask what was pasted (draft text, customer material, code, personal data), over what period, and whether the account was a personal or a work one. In the pack's example the four users confirmed the same day that no personal data and no Confidential files had gone in.

• Then: Do the answers show special category data or a never-allowed use?

Decision

Do the answers show special category data or a never-allowed use?

The users' answers can show what the first look did not. If special category data went in, or the use matches a prohibited practice, treat it as a stop now. Judgement: if the users cannot say what went in, answer yes.

• Yes, or not known: Stop it now and tell Legal or the DPO the same day
• No: Did Internal, Confidential or personal data go in?

Decision

Did Internal, Confidential or personal data go in?

Internal or Confidential information that went into a tool not approved for it is reported as an event, and so is personal data that went into a tool not approved for it, even inside information that is otherwise Public (BC-5). Judgement: if the users cannot say what went in, answer yes. Public information with no personal data in it goes in under the Public rule, not as mishandling.

• Yes, or not known: Report it as an event through the incident route
• No, only Public: Is it an AI feature a supplier added to software we already use?

Step

Report it as an event through the incident route

This is an event report. The incident pack screens every report and decides whether it becomes an incident and whether any reporting duty and clock applies, so set no clock here. In the pack's example the free assistant held only draft campaign wording (Internal). Triage found no incident the same day and no reporting duty arose.

• Then: Is it an AI feature a supplier added to software we already use?

Decision

Is it an AI feature a supplier added to software we already use?

Suppliers add AI features in releases. The pack treats the feature as a new use and screens it. If it is already switched on, it is a use found running, and a missing supplier decision becomes a dated condition, as it did for the pack's warehouse forecasting module. A separate AI tool, or a free assistant, is a different case: answer no. Notice of a new AI feature is supplier term ST-7, which P06 clause CL-20 carries.

• Yes: Screen it as a new use, marked already running
• No: Stop the use within 3 working days and help people move

Step

Screen it as a new use, marked already running

Answer the twelve intake questions as the use runs today and mark it as already running. The screen sets the tier as for a new use, taking the highest tier any trigger gives. A hold that would stop a new use becomes a condition with a date, because stopping a service people rely on may do more harm than carrying on for a short time under control.

• Then: Does the screen set Tier 2 or Tier 3?

Decision

Does the screen set Tier 2 or Tier 3?

Tier 2 applies when Confidential or personal data go in, customers or the public see the output, the use informs a decision about a person, or a wrong output would affect a core service. Tier 3 applies when the use falls in an Annex III area, decides or shapes a decision about a person, or acts on its own. Legal confirms any Annex III or role question. A provider signal, such as a feature built on a supplier's model API, does not raise a running use to Tier 3. H-4 becomes a dated condition until Legal records the role. A deployer role clears it and the answers set the tier. A provider or other role clears it but the use is then Tier 3, outside the pack's deployer governance. Else Tier 1.

• Yes: Approve with dated conditions, or stop it with a reason
• No: Decide: approve with conditions, or stop it with a reason

Outcome

Approve with dated conditions, or stop it with a reason

The tier's approvers approve with conditions (the executive decision-maker for Tier 3), or the use is recorded as Stopped with its reason. The use owner is consulted and signs the conditions it accepts, and nobody approves a use they own (IS-4). Write each condition with a date and the person who answers for it, and say in the decision that the tool is switched off if a date is missed. The assessment is due within 90 calendar days of the inventory closing by default, and the date goes in the inventory. Judgement: if that date has passed when you find the use, the approvers set a new date. The pack does not say.

Outcome

Decide: approve with conditions, or stop it with a reason

The AI governance lead decides a Tier 1 use, by default within 5 working days of a complete request (the pack's example took 4). If the lead owns the use, the executive decision-maker decides in the lead's place (RS-16). If it goes on, enter the use, tier, approver, conditions and review date in the inventory, and add the tool to the approved list. Tier 1 is reviewed at 12 months by default. If it cannot go on, record it as Stopped with its reason.

Step

Stop the use within 3 working days and help people move

A tool that is not on the approved list, or a use that needs data above what the list allows, is stopped within the working days the policy sets, 3 by default. The help matters as much as the stop: people who have an approved tool stop using the free one. Ask what they would need from an approved tool.

• Then: Do people carry on after being told to stop?

Decision

Do people carry on after being told to stop?

Nobody is disciplined for a mistake reported at once. The policy says deliberately using an unapproved tool after being told to stop, putting Confidential information into a personal or free account, hiding a use, or putting special category data into any tool may lead to action under the disciplinary procedure.

• Yes: Take it to the line manager and HR as a conduct matter
• No: Does an approved tool already meet the need?

Outcome

Take it to the line manager and HR as a conduct matter

This is a conduct matter for the person's manager and HR, under the disciplinary procedure the policy names. Judgement: pass on the dates the person was told, and keep it apart from the no-blame reporting of the original event, so that reporting stays safe for everyone else. The pack does not set either step.

Decision

Does an approved tool already meet the need?

Compare what the users wanted with the approved list. In the pack's example the four users moved to the approved assistant and finished the AI essentials module by 28 Sep, a week after the stop. If the list has nothing that meets the need, or you do not know, answer no.

• Yes: Move people to the approved tool and count the stop
• No, or not known: Propose the tool they need through intake

Outcome

Move people to the approved tool and count the stop

Move the users. The policy's evidence for a stop is the date found, the date stopped and the tool they moved to. Count it in the measure of unapproved uses found and stopped in time, such as 1 of 1, with any gap named, and never as a score.

Outcome

Propose the tool they need through intake

If no approved tool meets the need, the use owner proposes one through intake. Ask what they would need from an approved tool, then approve one. Judgement: refusing without an alternative is likely to send the use back to personal accounts. Until a decision is made, only Public information goes in, and never through a work account.