What it means
An AI risk assessment is a structured review of what could go wrong with an AI system, who could be harmed, how likely and serious that would be, and which measures reduce the risks. It considers technical, legal, ethical and organisational risks before and during use.
Real-world example
Before deploying an AI tool that drafts responses to citizen complaints, a city council assesses risks such as factual errors, inappropriate tone towards vulnerable people, disclosure of personal data and over-reliance by staff. It decides that all drafts must be reviewed, sensitive complaint categories are excluded, and error rates will be reviewed monthly.
Why it matters for AI adoption
Risk assessment turns general concerns into specific, manageable decisions. It is also the basis for meeting legal requirements such as GDPR data protection impact assessments and, for certain public-sector high-risk uses, fundamental rights impact assessments under the EU AI Act. Assessments should be revisited when the system or its use changes.
Questions to ask
- Who could be harmed if this system fails, and how?
- Which risks have we accepted, which have we reduced, and who signed off?