#21298 intentionally restored our original behavior of returning proxy class methods from AopUtils.getMostSpecificMethod, as it was pre 5.0.5. If you're seeing a regression with AspectJ pointcuts there, let's rather try to address this at the AspectJ evaluation level. Could you provide a test case for a corresponding AspectJ pointcut? Did your scenario work against 5.0.4 before or just against 5.0.5, i.e. is this actually a regression against 5.0.4 or just against a 5.0.5 side effect?