Stale data is data that no longer reflects the current state of its source or the conditions it describes, because nobody updated it after a change. A cached price that the source system has since revised, a customer address that moved, or a policy summary written before a rule was amended are all stale. Staleness is usually measured against an expected update interval: data older than that interval is treated as out of date.
Age is only one way data goes stale. A document can be the newest version in a knowledge base and still contain a rule that has ended, such as a temporary exception that expired last month. Freshness checks would pass, while an AI system that applies the rule today would give a wrong answer. For AI work, it helps to record when data was last updated and also how long each statement inside it remains valid.