AI safety incident crisis communication
When an AI system causes a safety or privacy incident, the response window is short and the communications are high-stakes: a clumsy customer notice, a premature press line, or a missed regulator deadline can compound the original harm. This tool generates structured draft communications for the audiences that matter — internal staff, customers, regulators, and press — tailored to the incident type, who is affected, and the severity, so you start from a sensible scaffold instead of a blank page under pressure.
How it works
You describe the incident: the type (harmful output, data exposure, an erroneous automated decision, a service outage, or a security breach), the affected parties (customers, staff, the public, a specific group), and the severity. For each audience you select, the tool assembles a draft with the right structure and tone — an internal notice that mobilises the team, a customer notification that states what happened and what to do, a regulator disclosure that hits the expected elements, and a holding press statement. Bracketed placeholders mark the incident-specific facts you must supply, and the drafts deliberately avoid speculation and over-commitment.
Tips and notes
- Verify before you assert. Say what you know; do not commit to facts you have not confirmed. “We are investigating” beats a wrong claim.
- Mind the regulator clock. Some privacy regimes require notice within 72 hours — confirm your deadline at the very start.
- One source of truth. Align internal, customer, regulator, and press messaging so they do not contradict each other.
- Always route through legal and comms. These drafts are a head start, not a substitute for review before release.