Incidents and notification clocks
An incident is a record of something that happened and needs tracking: a data leak, a prompt injection, an agent that acted wrongly, an unapproved tool or a vendor incident. It has an owner, a status,…
Written By Christopher Scaminaci
Last updated 4 days ago
An incident is a record of something that happened and needs tracking: a data leak, a prompt injection, an agent that acted wrongly, an unapproved tool or a vendor incident. It has an owner, a status, a timeline, links to findings and clients, and a review that closes it. Some incidents carry a legal deadline, called a notification clock. This page covers how to open an incident, work it and close it.
At a glance
- What it does: tracks serious events from the first report to the post-incident review, and tracks legal deadlines.
- Plan: Business and above, to open one or to let a rule open one. Any role can read the register. After a downgrade, existing incidents stay readable.
- Who: Administrator or Reviewer opens and works incidents. Any governance role reads them.
- Where: AI Governance → Alerts → Incidents.
- Needs: nothing to start. A person opens an incident by hand, or a rule opens one.
Open an incident
An incident is opened by a person, by a detection rule whose action is Open an incident, or by an incident rule you write. See Incident rules.
- Open Alerts → Incidents and select Open an incident at the top of the page.
- In the Open an incident panel, enter a short title and choose the type: Data leak, Prompt injection, Agent misaction, Unapproved tool or Vendor incident.
- Choose the severity (Info, Low, Medium, High or Critical).
- Describe what happened. Link findings if they relate. Set when your organization became aware. A time you can correct, never in the future.
- Select Open incident.
You become the owner. The register has no AI summary: every word is typed by a person or comes from the system.
A rule that fires very often opens at most a set number of incidents per organization per day. It then collects the rest in one overflow incident, so a noisy rule cannot flood the register.
Incidents that open on their own
Two things open incidents for you. A data-loss rule that is Live and whose action is Open an incident collects its findings into one incident a day. An incident rule works on findings of any rule, on alerts and on AI activity, with filters, a count threshold and your choice of adding to the open incident or opening a new one each time. Incident rules are on Alerts → Incident rules.
Work an incident
An incident moves through these statuses, in order: Open, Investigating, Contained, Resolved, Closed. Closed is reached only through the post-incident review.
On the incident's page you can:
- change the status and owner, and Save changes,
- Add note to the timeline,
- Add link to a finding or a client, or Remove a link,
- record decisions on its clocks,
- finish with Complete the review and close.
Closing an incident needs its post-incident review to be completed.
Notification clocks
Some incidents carry a legal deadline. A clock runs from the moment your organization became aware of the incident. StackJack alerts the owner once when a quarter of its period remains, and once more when it runs out.
The clocks StackJack offers are:
- GDPR notice to the authority (72 hours)
- NIS2 early warning (24 hours)
- NIS2 incident notification (72 hours)
You can mark a clock met or not applicable with Record decision. For not applicable, choose a reason:
- The incident is not significant
- The organization is not a NIS2 entity
- Another reason (say which in the note)
A clock that ran out is marked "missed" and stays on the record. StackJack records the deadlines. Deciding whether one applies is your organization's job.
What you will see
- A register of incidents, newest first.
- "Incident opened" and "Incident clock due" alerts in the inbox. The starting email policy "Incidents and approvals" sends them.
- Incidents in the insurer pack and, where an incident involved a client, the per-client pack.
- A line on the Overview when incidents need attention.
From your AI tool
A reviewer or administrator can open an incident from an AI tool. The tool needs the Business plan, as the Portal does. See Use AI Governance from your AI. Closing an incident stays in the Portal.
Troubleshooting
Next
Was this helpful?
More in Control and respond
Rule packs, modes and custom rulesFindings and triageData-loss rules, findings and scanning StackJack tool trafficFilter personal data out of tool results (the PII filter)Still need help? Ask the team