About halfway through this blog post the topic of LLM "personas" is mentioned, and while the given examples are mostly just jokes, the actual use case for that would solve the problem you perceive of this UX, in that you could simply use a persona that is more concise.

But to answer the "why?" - likely because this type of AI is a tool that is going to become universal very rapidly, and the overwhelming majority of the population has little to no understanding of the finer details of the technology. So the default persona has been fine-tuned to include the annoying "as an AI language model [...]" disclaimer very frequently, amongst other disambiguations, couched remarks, and clarifications that a power user would not necessarily require.

Conversational UIs allow users to use a language they already know well.

You realize that blabbering makes OpenAI more money right?