There was this Rocko thing a while back (which is not supposed to be discussed), where if I understood that nonsense correctly, the idea was that the decision theories here would do equivalent to one-boxing on Newcomb with transparent boxes where you could see there is no million, when there's no million. (and where the boxes were made and sealed before you were born). It's not easy to one-box rationally.

Also in practice usually being simulated correctly is awesome for getting scammed (agents tend to face adversaries rather than crazed beneficiaries).