If the data is there, it can be extracted, because the intended recipient was bound to have some way of extracting it. The question is how to know it's there and what to look for. For instance if you know the data is encoded in the LSB of a certain set of pixels in a lossless image, you can pull that data without the original. If you know roughly what to look for, like the data being in the LSB but aren't sure where, it's possible to run various types of pattern recognition on it. Plaintext encoded in the LSBs would be super obvious, for instance. But so would "noise" in an area of the image where LSB noise was expected to be low.

I have no idea what the state of the art is in steganographic detection, but I'm sure it involves the use of statistical tools to identify unexpected patterns in the data. Even if you used cryptography to encode a small piece of text so it became bitwise gibberish and then introduced it into a file through steganography, analytics could probably determine that the entropy of a particular part of the file shot way up. Once the data is found, the problem is no different than deciphering the intercepted message. Although steganography adds a layer of difficulty to the problem, it's just one layer. And if the trick is ever discovered, it stops being useful (to you) forever; you just have to find a new way of hiding data.