Skip to content
Post Mortem

Your human-approved AI incident squad

Not autonomous remediation. A set of helper roles that speed up the work while people keep every important decision.

Why a squad, and why human-approved

AI helpers are good at reading lots of records quickly, drafting summaries, and suggesting next steps. They are not good at knowing what your business can afford to break. So the squad proposes, and a person decides.

That keeps the speed benefit without handing an automated system the power to make an incident worse.

The roles

Sorting helper

Groups the alerts, drafts a first timeline and works out what is affected.

Needs
Look-only access to alerts and activity records.
Must never
Change anything.

Evidence helper

Collects records and evidence, and lists the accounts and systems that were affected.

Needs
Look-only access to activity records.
Must never
Delete or change evidence.

Stop-the-damage advisor

Suggests safe ways to stop the problem spreading. A person approves anything that deletes or breaks things.

Needs
Look-only access, plus the ability to draft actions for a person to approve.
Must never
Run destructive actions on its own.

Message drafter

Drafts updates for customers, colleagues and partners.

Needs
The confirmed facts and the timeline.
Must never
Publish without a person approving the wording.

Review writer

Writes a no-blame timeline, suggests what caused it and proposes next steps.

Needs
The timeline, activity records and chat history.
Must never
Blame individual people.

Practice-attack advisor

Suggests realistic tests to run on a safe practice copy.

Needs
A practice copy only.
Must never
Touch the live system.

Principles

  • AI helpers suggest. People approve anything that could cause big damage.
  • Everything an AI helper does is recorded.
  • Each AI helper has only the access its job needs.
  • Sensitive actions need a person to say yes.
  • An AI helper can be paused or cut off quickly.
  • AI helpers never hold unlimited keys to your live systems.
  • AI helpers keep evidence instead of deleting it.

How the first 30 minutes can run

  1. 0 to 5 min

    A person declares a problem and names who decides. The sorting helper drafts what is known. Stop the damage first: cancel the exposed key, pause anything suspicious that runs by itself.

  2. 5 to 15 min

    The evidence helper lists affected accounts and systems. The person in charge decides how to stop the damage. The advisor suggests, a person approves.

  3. 15 to 30 min

    The message drafter writes a short, factual update. A person edits and approves it. Everything is logged in one timeline.

  4. After

    The review writer drafts a no-blame review. The team edits it and agrees on a few real actions.

Can your team survive the next 30 minutes?

  1. 01Could you cancel a leaked secret key in under five minutes?
  2. 02Do you know which AI tools can reach which of your accounts and data?
  3. 03Do you know where your secret keys and passwords are kept?
  4. 04If one account were taken over, do you know what else it could reach?
  5. 05If a change to your website goes wrong, can you undo it?
  6. 06Do you know who would talk to customers if something went wrong?
  7. 07Is there a place where you would write down what happened and what you decided?
  8. 08Are the add-ons and borrowed code in your app locked to known versions and checked for flaws?
  9. 09Could you pause or cut off an AI helper in minutes?
  10. 10Have you practised what to do in a security problem in the last three months?

Answer all 10 questions to see your profile. Nothing leaves your device.