Clarifications:

Does it actually have to play a game or can we derive this being true from its ELO score against say Stockfish?
I assume this is considered "true" already for Deepmind due to MuZero?

AIs in 2028 will probably be optimized for planning and accomplishing common economically useful real-world tasks, but beating a super grandmaster at chess is pretty specialized. I would guess that even a very smart human - say, Leonardo da Vinci or Terence Tao - would probably have to work quite hard to attain that level of chess-playing, so I doubt that an AI could do it without at least some specialized fine-tuning.

Also, I'm not sure that a super grandmaster would try playing one of these generalized AIs by 2028, even if it had the capability to beat them.

If someone designs a new chess AI like that, some top streamer like Hikaru would definitely try playing against it.

@CDBiddulph mu zero presumably would qualify here. This has already been achieved by other labs.

Is it allowed to utilise other AIs like how some LLMs solve math with Wolthram Mathematica or must the entire process be OpenAI?

@TheAllMemeingEye yes the system's shouldbbe entirely OpenAI

An LLM or a chess AI such as alphago ?

@Odoacre any AI not made specifically and only for chess.