Pillow EUVD-2026-104598

Requires: (1) the optional olefile package installed alongside Pillow (not a default Pillow dependency); (2) Pillow version 7.2.0 up to but not including 13.0.0; (3) the application reading n_frames or saving all frames of an attacker-controlled MIC file; and (4) that MIC file embedding a TIFF with at least three IFDs so MicImageFile.tell/seek frame state never advances. Limiting factors: exploitation is impossible without olefile installed and without the app invoking the frame-count/save-all code path, so default Pillow installs that never touch MIC frames are not affected; impact is confined to CPU/availability exhaustion with no data exposure or code execution.

This is a denial-of-service issue (CWE-835, uncontrolled infinite loop) with no confidentiality or integrity impact - the provided CVSS 4.0 vector (AV:N/AC:L/AT:P/PR:N/UI:N/VC:N/VI:N/VA:H) scores 8.2 solely on high availability impact (VA:H), driving a CPU-pinning worker hang. The most important modifier is AT:P (Attack Requirements Present): exploitation is NOT universal because it requires the optional olefile package to be installed AND the application to invoke n_frames or save-all on an attacker-supplied MIC file. No EPSS score, KEV listing, or SSVC decision data was supplied, so exploitation-probability signals are absent; there is no public exploit identified (the only GitHub code artifact is the fix commit/test, not weaponized PoC). Signals are internally consistent - network-reachable but conditioned on a non-default dependency and a specific API call path - so this is a real availability concern for services that process untrusted images with olefile present, but it is materially lower priority than the 8.2 suggests for the majority of Pillow deployments that do not have olefile installed or do not read MIC frame counts. Note the input CVSS 4.0 AT:P correctly reflects the olefile/MIC precondition; a naive 3.1 reading as AV:N/AC:L/PR:N would overstate ease of attack.

An attacker uploads a crafted MIC (OLE-wrapped) file containing an embedded TIFF with at least three IFDs to a Python web service that uses Pillow with olefile installed to inspect or re-save image frames. When the service calls n_frames or save-all on the file, TIFF frame traversal never advances and the worker process spins at 100% CPU indefinitely, degrading or denying service. No public exploit code is identified at time of analysis; the vector is network-reachable and low-complexity but gated by the olefile dependency and specific frame-handling call path.