Possibly, but I think it may depend on the model. This scenario seems more likely for statistical models, especially those based on analogs, which may not handle extrapolation to a new climate state as well as it handles interpolation within the climate state in which the model was trained. For our more sophisticated global climate models, the intention is to use equations that capture the essential physics that are not dependent on climate regime. Of course, climate models cannot resolve all physical processes explicitly, and so they use parameterizations that are tuned based on relationships in the current and past climate. Therefore, I could see some risk that such parameterizations work better in the climate regimes for which they were tuned, but I am not sure if that would lead to a significant degradation of forecasting ENSO teleconnections in a different global climate regime.