Earlier quoted context omitted.
If the software to watermark is widely available (as it appears to be) then an adversary has all they need to corrupt any existing watermark. These steganographic watermarks depend on no knowledge of the process. If the method is particularly ingenious (one of the inputs is centrally stored entropy which the extractor references by trialing them all) then knowledge of the process alone may not be sufficient to obtain…
> If you obtain two or more copies of the decrypted content you will be able to diff them and work out what you need to corrupt even without knowledge of the watermarking process. By the time you've destroyed enough of the signal to remove the watermark, the content is unwatchable.
How would one know this? Care to share with the class?