The trigger
A staff member reports that a tool exposed customer data, a vendor emails that it is investigating an incident, or a complaint escalation hints at something wider. Each can be the first sign of a matter with a statutory clock attached, and a third party's breach can start your clock from the moment your organisation becomes aware. The first hour decides whether the regime owner receives evidence or hearsay.
The inputs
- The initial report exactly as received, with its timestamp
- The system, vendor or process involved and where its records and logs live
- Your incident intake checklist and evidence pack template
- The contract notification clause, if a vendor is involved
- The list of regime owners: privacy officer, breach reporting owner, complaints owner, whistleblower program owner
The steps
- Run the whistleblower gate first. If the matter could be a protected disclosure, stop: it never goes into a general AI tool. Hand it to an eligible recipient and end this recipe.
- Preserve the raw evidence in the approved incident system: the report as received, timestamps, prompts and outputs if AI was involved, and the location of any vendor or system logs. Build an evidence pack, not a bare ticket.
- Build a banded working record for AI use. Replace every name, date and system identifier with placeholders such as [DATE_BAND], [BUSINESS_AREA_BAND], [ALLEGATION_CATEGORY] and [SYSTEM_CATEGORY].
- Run the prompt below over the banded record in your approved AI tool. It produces a chronology separating confirmed facts, working assumptions and unresolved questions.
- Review the routing questions it returns on which regimes could be engaged: a privacy breach assessment, licensee breach reporting, or the complaints regime with its own enforceable clock. Treat them as a routing aid, never a conclusion.
- Route the pack to the owner of each candidate regime, record the handover time, and flag that the awareness moment may already have started a clock.
The prompts
Paste the banded record only. Never the raw report.
The checks
- Check every chronology entry against the source record before the pack moves
- Search the prompt and output for names, exact dates or system identifiers; if any appear, delete the chat and rebuild the banded record
- Confirm the output contains no reportability conclusion; if the tool drifted into one, strike it
- Confirm the awareness timestamp is recorded, because a clock may run from it
Governance controls
- The reportability assessment belongs to the human who owns the relevant regime. AI assembles and structures; it never classifies the matter against a statutory test and never drafts a regulator notification unreviewed.
- Banded placeholders are mandatory for a live incident. No raw names, dates or system identifiers enter any AI tool.
- The whistleblower gate is absolute: a matter that could be a protected disclosure goes to an eligible recipient with no AI involvement at any step.
- Record what the AI structured and what the person decided, and keep the raw evidence in the incident system.
- Hand any vendor contract notification clause to the regime owner in the same handover; the notice window shapes everything after.
The output
A one-page triage pack: a fact-checked chronology, a preserved evidence list, routing questions, and a recorded handover to the regime owner who makes the assessment.
Time it takes
About 60 minutes on the first run, most of it building the banded record carefully. A practised run with templates in place lands closer to 30.
TheAICommand. Intelligence, At Your Command.