FFmpeg CVE-2026-65706

Exploitation requires that the vulnerable FFmpeg build actually invoke the swaprect video filter on attacker-influenced input, and that the input be an NV12 (semi-planar two-byte interleaved chroma) frame with an odd width such as 17 pixels - even-width frames and non-NV12 pixel formats do not trigger the mismatched buffer size. Per the CVSS vector (AV:L, UI:P), the attack is local-vector and requires passive user or pipeline interaction to process the malicious frame; it is not remotely reachable without someone feeding the crafted content into the filter. The strongest limiting factor is that swaprect is a non-default, explicitly-configured filter, so environments that never use it are not exposed regardless of FFmpeg version.

The supplied CVSS 4.0 vector (AV:L/AC:L/AT:N/PR:N/UI:P/VC:H/VI:H/VA:H) scores 8.5 (High) and is internally consistent with the description: exploitation requires a local attack vector plus passive user interaction (a victim or pipeline processing a malicious frame), not remote unauthenticated access. The high impact metrics reflect heap corruption with potential code execution, but the description is explicit that a reliable, confirmed outcome is a process crash, with code execution characterized as 'potential' - heap-corruption-to-RCE is non-trivial and depends on allocator state and the host application. No EPSS score, KEV listing, or POC availability was provided, so there is no evidence of active exploitation; the realistic near-term risk is denial of service against video-processing workflows rather than mass exploitation. The tags label this RCE, but that is an upper-bound characterization, not a demonstrated capability.

An attacker crafts a malicious video (or NV12 frame stream) with a 17-pixel odd width and delivers it to a service or user whose FFmpeg-based pipeline applies the swaprect filter. When the frame is processed, the 18-byte chroma copy overflows the 17-byte heap buffer, corrupting adjacent heap metadata and crashing the process; with careful heap grooming the overflow could be steered toward code execution. No public POC was provided, so this scenario reflects the described mechanism rather than a demonstrated exploit.