Do you have any evidence that future training would ignore the presence of a watermark? Seems like a pretty valuable signal to me.
This is simply false. You are underestimating the utility of synthetic data and the ability to learn from the mistakes the current weights make.