You can't do anything about a compromised app or JS from a random website. I always find it weird when people attack Wayland's security model, more isolation is obviously a great idea, as demonstrated by supply chain attacks in the recent decade.

It's that Wayland's design, implementation, their attitude, and everything else about it is terrible. It could have been implemented without compromising on features or convenience by explicitly specifying minimalistic controlled side channels in their security model from the start, instead of shifting it onto ad-hoc implementations. And of course the windowing system is already too large of an attack surface. Many people are thinking about going full Qubes due to the current realities, while the others live in denial and call even window isolation "paranoia". Fascinating.