CVE-2026-111497 Brotli out-of-bounds write #10386

Two conditions are required: the application must pass attacker-controlled font data to ImageFont, and that data must be WOFF2.

Where have these conditions been met in public? Which applications or sites have been attacked using this path? I suspect anyone with an answer to that question has already patched their Pillow and would not benefit from a new release. If someone said "all versions of very popular app are vulnerable to this attack" then obviously we'd be doing a patch release, but I haven't seen anything like that, ever in 16 years of Pillow (that I can recall!)

That (open untrusted font) was one of our oss-fuzz paths that uncovered some font handling bugs.

If the position is that this vulnerability doesn't warrant an official patch release ahead of the the scheduled quarterly, and that downstream users ought to do a custom build on their own, that's fine.

Not intentionally (woff2 was not really expected to work here), but Weblate could be attacked via this path. We've closed it by filtering out the payload before passing it to Pillow in WeblateOrg/weblate#22584 (what should have done this anyway).

Even if this is fixed for us, it is still annoying that anything shipping Pillow will trigger security scanners because of the bundled vulnerable Brotli for the next two months.

Hi. In many scenarios it's really not that simple as "i'll just build it myself". I imagine most people don't use Pillow directly but it's rather installed by dependencies of software they build, during build pipelines. Now it is indeed showing in scanners for a very high profile CVE, and not having an update for this is surely causing a lot of work for some folks to keep compliance and security organizations happy. Not criticizing, just adding a perspective.