Skimming the PNAS paper, it appears that the conclusion is that evolved group co-operation is not mathematically stable without evolved altruistic punishment. I.e. populations with only evolved co-operation drift towards populations without any group focused evolved traits, but altruistic punishment seems to exclude enough defectors that evolved co-operation maintained frequency in the population.

Which makes sense, but I'm nowhere close to qualified to judge the quality of the paper or its implications for evolutionary theory.