Training data access wouldn't be that relevant, this would be fed to "Sydney" as a preface to the "conversation" so it's more like a part of the bot source code

Training data access wouldn't be that relevant, this would be fed to "Sydney" as a preface to the "conversation" so it's more like a part of the bot source code and not the LLM itself. But yes, it's probably not possible to talk with certainty whether it's real or a hallucination. The fact that it seems consistent across wildly differing ways of querying it gives us some reason to believe it's more than just a hallucination thought.