FFmpeg EUVD-2026-62584

Exploitation requires that FFmpeg be invoked with the RTP output muxer ('-f rtp') on an input stream containing AV1 video (codec av1), specifically triggering the keyframe detection logic in libavformat/rtpenc_av1.c. The crafted AV1 packet must contain an OBU with a payload size field large enough to cause the remaining-size counter to underflow and wrap on subtraction. The CVSS vector AV:L/UI:P confirms that the attack surface is local processing of attacker-supplied content - an operator or automated process must initiate the FFmpeg mux operation on the malicious file or stream. Exploitation is constrained by the requirement that the AV1-RTP code path is actively exercised; deployments using FFmpeg only for local transcoding, file-to-file conversion without RTP output, or for codecs other than AV1 are not reachable. Remote network-based exploitation without local file access or pipeline injection is not supported by the provided CVSS vector.

The CVSS 4.0 base score of 6.9 (Medium) is backed by a vector of AV:L/AC:L/AT:N/PR:N/UI:P with VC:H and VA:H. The local attack vector (AV:L) and passive user interaction (UI:P) are the primary risk mitigators: exploitation requires that FFmpeg be used to mux a crafted AV1 file to RTP output, which is not a universally deployed code path. Real-world risk is therefore narrower than the high confidentiality and availability impacts suggest in isolation. No EPSS data was provided in the available intelligence; the vulnerability is absent from CISA KEV and no POC code has been publicly disclosed. The risk is most acute in automated media transcoding pipelines, live-streaming infrastructure, or video conferencing back-ends that accept untrusted AV1 content and route it through the RTP muxer without sanitizing OBU length fields upstream. In those contexts an attacker supplying a crafted AV1 stream could force memory disclosure (potential information leak) or reliably crash the FFmpeg worker process. For deployments that do not use the AV1-RTP muxing path, practical risk is negligible. No conflicting signals are present: description, CVSS vector, and CWE are internally consistent.

An attacker operating a media upload service, or submitting content to a streaming back-end that internally muxes AV1 video to RTP, crafts an AV1 bitstream containing an OBU with an encoded payload size deliberately set to near the maximum representable value. When FFmpeg's keyframe detection loop processes the packet, the unbounded size arithmetic causes the remaining-byte counter to wrap, and the loop dereferences memory well past the end of the packet buffer on the next iteration. Depending on what occupies adjacent heap memory, this yields either a crash (denial of service against the FFmpeg worker) or an out-of-bounds read that returns heap contents - potentially including keys, tokens, or other sensitive data from the same process - back through the application's error path or logging. No public POC has been identified at time of analysis.