Skip to content
World Model Readiness
Engraved governance instrument

For your company

Governance & Compliance

Module · The first 24 hours decide it

The AI Crisis Communication Check

When an AI failure becomes public, the damage is set less by the failure than by how you respond in the first day. Silence reads as guilt, a bad statement reads as evasion, and both are hard to walk back. This module checks the five things that decide whether you can speak fast and well: prepared statements, a ready spokesperson, an escalation path to comms, clarity on who you must notify, and the discipline to hold one honest story.

Question 1 of 5 · Statements are drafted

If an AI system failed publicly tomorrow, is there anything already written to work from?

The worst time to draft your first public statement is while the story is breaking and the lawyers and executives are arguing. Holding statements for the plausible AI failure scenarios, written calm and reviewed in advance, turn hours of panic into minutes of editing. A blank page at hour one is how bad statements get made.

Question 2 of 5 · A spokesperson is ready

Is there a named person who could credibly speak about an AI failure on short notice?

Explaining an AI failure needs someone who can be both accurate and human, under questions from people who assume the worst. That is a specific skill, and the person who has it should be known and rehearsed before the crisis, not chosen during it. An unprepared spokesperson can turn a contained problem into the story.

Question 3 of 5 · Escalation reaches comms fast

When an AI problem is spotted, does it reach the people who handle communications quickly?

AI failures often surface first with an engineer or a support agent, far from anyone thinking about public response. If there is no fast path from there to comms and leadership, you learn about the crisis from a journalist. The clock starts when the problem is spotted, not when comms hears about it.

Question 4 of 5 · Notification duties are clear

Do you know who you are legally required to notify, and how fast, after an AI incident?

An AI failure can trigger hard deadlines: data breach notifications in hours, regulatory reports, contractual duties to customers. Missing a notification window adds a compliance failure on top of the original incident. These clocks are known in advance or missed in the moment.

Question 5 of 5 · One story, held straight

Could you keep one honest, consistent account across the whole incident?

Under pressure, the story fractures: engineering says one thing, sales another, and an early claim that turns out wrong poisons everything that follows. Post-incident discipline means one accurate narrative, corrected as facts firm up, owned by someone. The cover-up, or the contradiction, is usually worse than the failure.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

How prepared are you to communicate about an AI failure?

Is there a named person ready to speak about an AI incident?

Have you ever had to communicate publicly about an AI problem?

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Statements are drafted

  1. 1Nothing prepared
  2. 2General crisis plan
  3. 3AI mentioned vaguely
  4. 4Draft holding statements
  5. 5Scenario-specific drafts

At the low end: With nothing written, your first public words get drafted under maximum pressure and minimum time. Draft holding statements for two or three plausible AI failure scenarios while nobody is panicking. What good looks like: Scenario-specific drafts mean hour one is spent editing, not authoring. Review them as your AI use changes; a holding statement about a system you no longer run is a false start.

A spokesperson is ready

  1. 1No one identified
  2. 2Whoever is around
  3. 3Named, unprepared
  4. 4Named and briefed
  5. 5Named and media-trained

At the low end: Without a named spokesperson, the crisis chooses one for you, usually badly. Name who speaks for the company on an AI failure, and make sure they understand the systems well enough to be credible. What good looks like: A named, media-trained spokesperson is a genuine asset in the first hour. Keep them briefed on your current AI systems; credibility collapses fast when the spokesperson does not understand what failed.

Escalation reaches comms fast

  1. 1No path
  2. 2Word of mouth
  3. 3Path exists, unknown
  4. 4Known escalation path
  5. 5Fast, tested path

At the low end: With no escalation path, hours can pass between an engineer noticing and comms finding out, and those are the hours that matter. Define who a spotted AI problem goes to, and how, before you need it. What good looks like: A fast, tested escalation path means comms is working the problem while it is still small. Drill it occasionally; an escalation route that has never been used tends to break at the first real test.

Notification duties are clear

  1. 1No idea
  2. 2Assume none
  3. 3Roughly aware
  4. 4Documented per scenario
  5. 5Documented with timelines

At the low end: Not knowing your notification duties means an AI incident can quietly become a regulatory breach on top of the original problem. Map who must be told, and within what window, for your likely incident types. What good looks like: Documented notification duties with timelines mean the legal clock is managed, not discovered. Keep them current as regulations and contracts change; notification windows tighten more often than they loosen.

One story, held straight

  1. 1No coordination
  2. 2Everyone improvises
  3. 3Loosely aligned
  4. 4One owned narrative
  5. 5One narrative, disciplined

At the low end: Without a single owned account, different parts of the company will contradict each other in public, and the inconsistency becomes the story. Put one person in charge of the narrative and route all statements through them. What good looks like: A single, disciplined narrative that updates honestly as facts emerge is what rebuilds trust after a failure. Rehearse the discipline; the instinct to minimise or over-promise is strongest exactly when it does the most damage.