ImageMagick EUVD-2026-109545

Requires that an administrator has configured policy.xml to block specific delegates or executables - if no such policy restriction is in place there is nothing to bypass, so the flaw only has impact where policy.xml is relied on as the security boundary. The attacker must also get ImageMagick to process input that triggers a delegate invocation (UI:P: a user or automated process must feed the crafted input), and the vector is local (AV:L), not remote network-level per the supplied CVSS. The outcome is limited to running a program that is already present but policy-blocked (integrity-only, VI:L); it is not described as arbitrary remote code introduction. Limiting factors: dependence on a non-default hardening configuration, required input processing/interaction, and local vector all narrow realistic exposure below what the medium score alone might suggest.

This is a medium-severity authorization/policy bypass, scored CVSS 4.0 base 4.8 by the source (AV:L/AC:L/AT:N/PR:N/UI:P with only VI:L impact). The signals are internally consistent and point to a defense-in-depth weakness rather than a direct compromise: a local attack vector, required passive user/process interaction (UI:P), and limited integrity impact only (confidentiality and availability are N). The vuln.today decision label is 'this_month' (not act_now or emergency), which matches the picture of a bounded, non-emergency issue. No EPSS exploitation-probability score was supplied in the input, and the CVE is not listed in CISA KEV, so exploitation status is 'no public exploit identified at time of analysis' - the two references are vendor/VulnCheck advisories, not PoC or exploit code. The practical risk is conditional: it only matters to organizations that actually rely on ImageMagick's policy.xml to block specific delegates/executables as a security boundary; where that boundary is the main control preventing dangerous delegate execution, this bypass defeats it. Missing data to flag explicitly: no EPSS score is provided, so widespread-exploitation likelihood cannot be quantified, and the exact delegate-invocation parsing trigger is not detailed in the available sources.

An application uses ImageMagick to process attacker-supplied files, and an administrator has hardened policy.xml to block certain delegates/executables. The attacker submits input crafted so that ImageMagick's incorrect parsing of the delegate invocation causes the policy-blocked program to run anyway, defeating the administrator's restriction. Consistent with the local vector and UI:P, this requires the target to process the attacker's input; no public exploit is identified at time of analysis.