This isn't a failure of AI, but an example where context in your prompt/RAG/overall architecture is key to a good outcome if you understand the other information the system is trained on.