Is causal inference really possible, even in controlled experiments?

Causal inference has always been a topic in statistics, and most people would buy it if it is designed carefully. It is true that controlled experiment is probably the best we can do towards it. However, even with that, we cannot really get the outcomes simultaneously for both test and control treatments. We still rely on the BELIEF that two groups are similar in all aspects except for the treatment, as a result of randomized assignment. Are we really sure? Maybe a "deep association" is a better name, instead of "causation".