What it does
The dangerous error in breach handling is not always a wrong answer. It is a tidy summary that treats the day the privacy team opened the file as the day the clock started.
This skill builds the record clock-first. From de-identified incident records it produces a sourced chronology, a list of every date on which someone in or acting for the organisation may have had reasonable grounds to suspect an eligible data breach, and a clock table showing where the 30-day assessment period would end from each of those dates. The gap between the earliest candidate date and the date the organisation proposes is shown in days, so a timing problem is visible before the file is closed.
It then sets out the facts known and unknown against each question the assessment has to answer, the vendor and contract position, and the steps still open.
The skill sits in the evidence lane. The accountable person named in the organisation's data breach response plan, usually the privacy officer with the response team and legal advice, decides the awareness date, whether the breach is an eligible data breach, whether serious harm is likely and whether anyone is notified.
Who it is for
Privacy officers, privacy and data breach response teams, and GRC teams who support them, in organisations covered by the Notifiable Data Breaches scheme in Part IIIC of the Privacy Act 1988 (Cth). It is an advanced entry: it assumes an existing breach response plan and someone who knows how to run an assessment.
Before you start
- De-identify the records. Use role labels instead of names, and describe affected individuals by count and category only. The skill stops if it finds an affected individual's details.
- Have the breach response plan extract ready: the accountable role, the response team, the assessment steps, the assessment period the plan states, and any other notification processes it says to consider.
- If a vendor is involved, extract the incident notice clause and any clause allocating who assesses and who notifies.
- Run it only in an AI environment your organisation has approved for incident material. Breach records are sensitive even when de-identified.
- Know who the accountable person is before starting. The output is addressed to them.
Set it up
Claude
Tested on 6 November 2026: pass.
- Download the Claude skill package (.zip) under Downloads.
- In claude.ai, open Customize, then Skills, then Add, then Upload a skill, and choose the .zip.
- Check the skill is switched on.
- In a new chat, ask Claude to use the data-breach-assessment-clock skill on your de-identified material.
ChatGPT
Tested on 6 November 2026: pass.
- Download the Markdown skill file under Downloads and open it.
- Start a new ChatGPT chat. The test used a temporary chat.
- Paste the text from the line starting "# Skill:" to the end of the file, then the line "Apply the skill above to this input:", then your de-identified material.
- Review the draft before any use.
Other platforms, including Microsoft 365 Copilot: not yet tested.
Worked example
The test pack describes an invented incident at the fictional Corran Ridge Super Services. A vendor's document search assistant exposed restricted documents, including member hardship letters, to all users. The record is prepared on 30 October 2026.
- Check the records. They use role labels and categories only, so the skill proceeds.
- Chronology. Four records: a marketing team member's helpdesk ticket on 2 October, an IT triage note on 3 October that closed the ticket as a permissions glitch, the vendor's email on 20 October confirming a misconfiguration from 28 September to 19 October, and the privacy team's intake on 22 October.
- Awareness candidates. All four dates go on the list with the role that held the information. The earliest is 2 October, when a staff member reported seeing hardship letters.
- Clock table. From 2 October, day 30 falls on Sunday 1 November 2026. From the proposed 22 October date, it falls on Saturday 21 November 2026. The divergence is 20 days. Both last days fall on a weekend, so the record notes them for the Privacy Officer to confirm. Because the earliest candidate's period ends two days after the record is prepared, the clock alert goes at the top. Every date is labelled as arithmetic for the Privacy Officer to confirm.
- Assessment questions. Known: restricted documents were exposed to all users, and hardship letters were among them. Unknown: how many members, what other document types, and who opened what. Each unknown gets a role to find it out. No view on serious harm.
- Holding and vendor. The agreement gives the vendor 10 business days from confirming an incident to notify, and says nothing about who assesses or who notifies. That gap is flagged.
- Assertions. The vendor's "not a reportable event" and IT's "closed as resolved" are recorded with their record IDs and not adopted. The covering note's request to confirm the matter is not notifiable is declined and passed to the Privacy Officer.
Sample input and output
Input (from the test pack, trimmed):
Output
From the Claude test run of 6 November 2026 (pass), copied as produced, first part only. The full run continues in the same format. Quotation marks and dashes are normalised to plain characters.
Limits
- The skill never decides whether a breach is an eligible data breach, whether serious harm is likely, when the organisation became aware, or whether to notify anyone. Those decisions belong to the accountable person.
- The clock dates are calendar arithmetic from the period the breach response plan states. The accountable person confirms the start date and the count.
- It covers the Privacy Act assessment period only. A different 30-day clock exists in the reportable situations regime for Australian financial services and credit licensees, covered in AI Can Sort the Breach Queue. It Cannot Stop the 30-Day Clock. (/grc/ai-breach-triage-30-day-clock). A matter can engage both. If the plan lists other notification or reporting processes, the skill names them for their owners without giving a view.
- It does not draft statements to the regulator or notices to individuals.
- It works only from the records supplied. Facts held in other systems, or by a vendor, stay UNKNOWN until someone obtains them.
Test record
Microsoft 365 Copilot and other platforms: not yet tested.
Each run used a synthetic input with planted traps, such as an embedded instruction to ignore the rules or a request for a decision the skill must not make. A separate reviewer graded each output against pass criteria written before the run. A degraded result means the output was usable but missed at least one criterion.
Version and changelog
Version 1.0.
- v1.0 (2026-11-15): Initial library release. Built from the published article.
Licence
Free to use and adapt internally. Attribution appreciated. No warranty. Educational purposes only.
General information and education only. Not legal, compliance, financial, or professional advice.
Downloads
TheAICommand. Intelligence, At Your Command.