PAC learning can be influenced by strategic manipulation, where data points change to get better results. The usual ERM principle doesn't work well in this situation. A new incentive-aware ERM principle has been developed with optimal sample complexity. Classifiers that prevent manipulation have a sample complexity that doesn't depend on the type of hypotheses used. In cases where manipulation follows a certain pattern, only using manipulation-proof classifiers is enough for good results. This shows that being manipulation-proof can help simplify PAC learning.