Requalification is the act of checking again, after a change, that a system or dataset still meets the criteria it originally met. The term is common in regulated engineering and pharmaceutical validation, where engineers requalify equipment after maintenance or relocation. The same logic applies to AI systems and the data they run on.
A change event triggers requalification; a fixed schedule does not. Typical triggers are new data, a new model or prompt version, a policy update, or a change in the runtime environment. For example, if a support team updates the data that a retrieval model reads from, the earlier qualification no longer describes the current state, so the team checks the affected task again under the same conditions as before.
Good requalification is scoped. It re-runs the original check only on the parts a change touched, compares the result with the previous one, and records the outcome as a new version. This keeps the evidence chain intact without repeating the whole validation.