Pillow EUVD-2026-104639

Requires that the target application apply a Pillow rank-based filter - ImageFilter.RankFilter, MedianFilter, MinFilter, or MaxFilter - to an image whose width or height is zero (sizes (0,1), (1,0), or (0,0) per the regression test). The attacker must be able to supply or cause creation of such a zero-size image reaching that filter call. Limiting factors: the application must actually use one of these specific rank filters (not a universal Pillow usage pattern), and the input path must allow a zero-dimension image to reach it - this is why the official CVSS 4.0 vector sets AT:P (attack requirements present) rather than treating it as unconditional. No authentication is indicated by the vector (PR:N), and no user interaction is required (UI:N), but impact is limited to a process crash (availability only).

This is a denial-of-service null pointer dereference (CWE-476) in the Pillow imaging library prior to 13.0.0. All signals point to availability-only impact: the provided CVSS 4.0 vector (CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:N/VI:N/VA:H) assesses only VA:H with no confidentiality or integrity impact, and CWE-476 plus the description ('native process crash') confirm a crash-only outcome rather than code execution. The score of 8.2 is driven by network reachability and high availability impact, but note the AT:P (Attack Requirements: Present) flag - exploitation is NOT unconditional; it depends on the target application actually applying a rank/median/min/max filter to an attacker-influenced image and on a zero-size image reaching that code path. No EPSS score, CISA KEV listing, or public exploit was provided in the input, so there is no evidence of active or at-scale exploitation. The GitHub source evidence provided is the upstream fix commit and a regression test - NOT a weaponized proof-of-concept. Bottom line: a credible, low-complexity DoS in an extremely popular library, but its real-world priority is moderated by the required application-side filter usage and the lack of confidentiality/integrity impact. Treat as a patch-in-normal-cycle availability bug rather than an emergency, unless the affected application exposes rank-filter processing to untrusted image input.

A web or batch application uses Pillow to process user-uploaded images and applies a rank-based filter (RankFilter, MedianFilter, MinFilter, or MaxFilter) as part of its pipeline. An attacker uploads or triggers creation of a zero-size image (width or height of 0); ImagingExpand() allocates output and then faults reading the first pixel from empty input storage, crashing the native process. No public exploit code was identified in the provided data, though the trigger is trivial to construct.