That's the coronagraph that blocks out the star itself so that everything else is visible.
https://blogs.nasa.gov/webb/2023/03/24/how-webbs-coronagraph...
https://blogs.nasa.gov/webb/2023/03/24/how-webbs-coronagraph...
Silly me for believing NASA.
Saturated pixels are not necessarily "invalid" the way cosmic ray bit flips and stuck pixels are, but setting them to zero does make the other bright but not saturated pixels much easier to identify.
When using a coronagraph, it's possible that that value is all over the place (diffraction around edges, noise, etc) but you know any data "under" the coronagraph is bogus and you know which pixels are covered, because you know how the device was made, so you can mark them as bogus yourself.
The actual value in the data isn't black (0,0,0), it'll be some obviously special value that won't otherwise happen, because black is a valid value.
I suppose the idea that not-misleading data trumps aesthetically-pleasing data continues then to the press release image, even though a solid white central spot would look more familiar to people who have accidentally included the sun in a photo.
>The 23.0 µm coronagraph of MIRI uses a classic Lyot mask to achieve high contrast imaging near bright sources. The mask is rather large, providing an inner working angle (IWA) of ∼ 3.3λ/D, i.e., ∼ 2. Our data is the first complete reference PSF subtracted set of observations taken with JWST using this mode.
There's some residual light and diffraction effects around the Lyot mask which they have manually deleted.