No; it won't corrupt them. It just (as expected) doesn't know how to diff them nor does it try to. The problem would be if they were massive and if they would change frequently (or sometimes more than once), which would mean the amount of data you'd need to 'git clone' would grow much more sharply than with non-binary files.
Consider games that have artists check in TIFFs that get processed during build process. Or perhaps pre-baked .zip files containing all game artwork for sake of faster game launch by developers. ("We don't need to build the .zip, just the .exe.")
Neither really applies to us, unless we start going crazy with icons.