No new code goes in that violates the rule, and ideally no code at all goes in that is both unsafe and parses untrusted data (regardless of sandboxing) and old code doing both gets replaced.
A giant pile of C++ can be used for rendering, not parsing untrusted data. A giant pile of C++ can sit behind a validator: a memory-safe JSON validator can vet a stream, before an C++ library deserializes it. Etc.
Mozilla's position for some time now has been, "we aren't opposed to shipping JXL support, but we'd want to ship a decent implementation in a memory safe language, not the reference C++ implementation". That position hasn't been met with very much criticism.
Google's position, on the other hand, has been a flat-out "no, we will not ship JXL". That's what has been met with criticism. Not an imagined reluctance to shipping a C++ JXL implementation.
It's Google, it's one of the biggest tech companies in the world making boatloads of money, in part off their browser. They're currently best known as one of the companies trying to create AI God. They really can't write an... image format parser?
