I think that we will see high level poker players take an approach similar to what Magnus Carlsen has done. He finds positions that have not been played frequently, if ever. That means they are outside some of the analysis done by machines. He also talks about how his team adds constraints to the engines they use like 'no castling' to see if they can find a local max from a position you would never actually play.
I am a very bad poker player, so I don't know what an equivalent strategy would look like, but we have had 'game theory optimal' type play for a while. I'm sure the best poker minds will think of clever ways to do things to confuse an AI player. Especially online when there are fewer tells.