>Out of interest, what's to stop them publishing some 'version' of the modified code, but actually running a different version on their servers?

The law

>Would that be easy to detect?

No