OpenImageIO CVE-2026-43903

Requires victim to process a malicious SGI format (.sgi, .rgb, .rgba, .bw, .int, .inta) image file using an application built against vulnerable OpenImageIO versions (pre-3.0.18.0 or 3.1.x before 3.1.13.0) in release build configuration. The SGI file must use RLE compression (not all SGI files do) with crafted count fields exceeding scanline dimensions. User interaction is mandatory (UI:A in CVSS) - the file must be explicitly opened or processed by the victim. Exploitation is LIMITED by requiring local file access (AV:L) rather than remote network delivery, though files can arrive via email attachments, cloud storage, or supply chain (malicious asset libraries). Defense is complicated by SGI format legitimacy in VFX workflows where blocking the format entirely disrupts production.

Real-world risk is MODERATE despite the 8.4 CVSS score. CVSS vector (AV:L/AC:L/PR:N/UI:A) indicates local attack requiring user interaction - attacker must deliver a malicious .sgi file and convince a user to open it, limiting remote exploitation scenarios. SSVC framework classifies this as POC exploitation (publicly available proof-of-concept exists), non-automatable (requires social engineering), with total technical impact (arbitrary code execution). However, no active exploitation confirmed (not in CISA KEV, EPSS data unavailable suggests low observed targeting). The attack surface is significant in VFX/animation studios where OpenImageIO processes untrusted image assets from external sources, vendors, or public repositories. Priority should be HIGH for organizations in media production pipelines, MEDIUM for general enterprise environments where SGI format processing is rare. The presence of POC code elevates risk from theoretical to practical, though the local+UI requirements prevent wormable exploitation.

Attacker crafts a malicious .sgi image file with RLE-compressed scanline data containing a count value exceeding the declared scanline width. VFX artist receives the file via email, shared storage, or asset repository and opens it in compositing software (Nuke, Houdro, Blender) or image viewer using OpenImageIO. The RLE decoder writes attacker-controlled pixel data beyond the heap buffer boundary, corrupting adjacent memory structures to achieve code execution with the privileges of the rendering application, typically the user's full desktop access in production environments.