By 2030, public-sector AI systems will need a published impact assessment before launch in more places than today

Prediction: by 2030, more jurisdictions than today will require government agencies to complete and publish an impact assessment before deploying an AI system that affects people's rights, benefits or services.

Reasoning: guidance on trustworthy and ethical AI from international organisations is already widely circulated, and governments are under pressure to show how automated decisions are checked. Procurement rules are an efficient place to add requirements, because they affect every vendor at once.

Confidence: moderate, about 65 percent that the number of places with such requirements rises. We are less sure that assessments will be detailed, independent or enforced.

What would prove us wrong: if the number of jurisdictions with such a requirement does not increase, or if assessments are required but rarely published.

How to check: compare lists of national and regional rules on public-sector AI in 2025 and 2030, and test whether published assessments exist for real systems.