When must AI systems be re-screened?
AI systems must be re-screened when the legal context changes, when the system itself changes, when the actor role shifts, or when the operational context evolves in ways that affect risk classification or obligations. The EU AI Act is implemented progressively — approved once does not mean compliant forever.
Re-screening is not an exceptional event. It is a regular governance activity that maintains the integrity of the original screening decision over time.
Key points
- Legal triggers include new delegated acts, harmonised standards, guidance documents, or amendments to the regulation itself.
- System triggers include version upgrades, changed use cases, new data sources, modified deployment context, or provider changes.
- Organisational triggers include actor-role changes (e.g., from deployer to provider through substantial modification) or new operational contexts.
- Re-screening must reference the original screening record, show what changed, and document the renewed decision with attribution.
- The re-screening result becomes a new versioned record, preserving the governance history rather than overwriting it.
Why it matters
An AI system that was compliant at initial screening may no longer be compliant if the regulation, the system, or the use case has changed. Without structured re-screening, the original governance record becomes stale and potentially misleading. Re-screening preserves the integrity of the compliance claim over time and ensures the organization can demonstrate continuous governance rather than point-in-time assessment.
How EAB approaches this
EAB's Re-Screening Queue monitors legal-change events and flags affected AI systems for renewed review. Legal Source Mapping anchors each governance record to the specific legal version consulted. Versioned Screening Records preserve the complete re-screening history, linking each new decision to its predecessor and the trigger that initiated it.