The date after which a model knows nothing, because its training data stopped there. Ask about events after that point and it will either say it does not know or, worse, confidently invent something. Many systems compensate by giving the model web search, but the underlying model itself remains frozen at its cutoff.
Technical view
Coverage thins out well before the nominal cutoff date, since recent material is under-represented at collection time relative to how much will eventually exist about it. Models are also often poor at judging what they do and do not know about the boundary period. Retrieval and search tools are the standard mitigation, which shifts the problem to the reliability of retrieval rather than removing it.
Related
Part of AI Terminology Index, section: Controlling and Extending Behaviour at Run Time.