FFmpeg CVE-2026-8461

Exploitation requires the target to decode an attacker-supplied MagicYUV-encoded stream through FFmpeg's libavcodec MagicYUV decoder (libavcodec/magicyuv.c) on a version before 8.1.2; the CVSS UI:R metric reflects that a user or automated pipeline must actually open or process the malicious file. Limiting factors: builds that have disabled the MagicYUV decoder at compile time are not vulnerable, ingest pipelines that whitelist a small set of codecs and reject MagicYUV cannot reach the vulnerable code path, and reliable code execution (versus DoS) further depends on the target's heap allocator, ASLR, and process mitigations. Server-side transcoders that accept arbitrary user uploads effectively satisfy the UI:R requirement automatically.

The CVSS 3.1 vector CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H scores 8.8 and reflects a network-reachable, low-complexity, unauthenticated attack requiring user interaction (opening or processing a malicious media file), with full CIA impact consistent with potential RCE. The 'denial-of-service and, in some cases, ... remote code execution' wording in the description is honest about uncertainty: reliable RCE from an OOB write depends on heap layout, ASLR, and mitigations, so DoS is the guaranteed outcome and RCE is opportunistic. EPSS, CISA KEV status, and POC availability were not provided in the input - absence of these signals is not evidence of safety, especially given the JFROG reporter and the public PR reference, which suggests a patch and likely details will surface quickly. Real-world risk is elevated for any service that automatically transcodes user-uploaded media (CDN ingest, video sharing, chat attachments), where 'user interaction' is effectively automated by the server.

An attacker crafts a malicious video file containing a malformed MagicYUV stream and delivers it via a video sharing site, chat attachment, email, or a URL that a media player or server-side transcoder will fetch. When the victim opens the file (or a backend transcoder automatically processes the upload), libavcodec's MagicYUV decoder writes out of bounds while parsing attacker-controlled fields, crashing the process at minimum and potentially executing attacker-controlled code with the privileges of the media-handling process. No public exploit identified at time of analysis, but the JFROG disclosure and public upstream PR make exploit development plausible.