CVE-2026-75147: FFmpeg Buffer Overflow Vulnerability

CVE-2026-75147 Overview

CVE-2026-75147 is an out-of-bounds read in the FFmpeg AV1 Real-time Transport Protocol (RTP) packetizer located in libavformat/rtpenc_av1.c. The keyframe detection loop that searches for a sequence header Open Bitstream Unit (OBU) advances its pointer and remaining-size counter without first bounding the OBU size against the remaining packet data. A crafted OBU size causes the remaining-size counter to wrap to a positive value, letting the next loop iteration dereference memory past the end of the packet buffer. The flaw affects FFmpeg builds prior to commit 983dae9 and is classified under CWE-125.

Critical Impact
A crafted AV1 input packet muxed to RTP triggers an out-of-bounds read that can leak adjacent process memory or crash the FFmpeg process.

Affected Products

FFmpeg versions built before commit 983dae9c19f46c87d597598c0fd2f2fcee0ad2f8
Applications that link libavformat and mux AV1 streams to RTP
Downstream media servers and transcoding pipelines embedding vulnerable FFmpeg builds

Discovery Timeline

2026-08-19 - CVE-2026-75147 published to the National Vulnerability Database (NVD)
2026-08-19 - Last updated in NVD database

Technical Details for CVE-2026-75147

Vulnerability Analysis

The vulnerability resides in the AV1 RTP packetizer in libavformat/rtpenc_av1.c. FFmpeg scans incoming AV1 packets for a sequence header OBU to detect keyframes before packetizing them for transport over RTP. The scan loop reads each OBU header, extracts the encoded payload size, and advances both the read pointer and a remaining-size counter by the header length plus the payload size.

The loop does not validate that the reported OBU payload size fits within the remaining buffer before performing the advance. A crafted payload size can exceed the remaining data, causing the arithmetic to wrap the remaining-size counter to a large positive value. On the next iteration, the loop treats stale memory beyond the packet buffer as valid OBU data and dereferences it, producing the out-of-bounds read.

Exploitation is triggered by a crafted AV1 input packet muxed into an RTP session. The attack complexity is low, but the vector is local and requires user interaction to process the malicious media, limiting large-scale remote exploitation. Public exploit code is not currently available, and the flaw is not listed in the CISA Known Exploited Vulnerabilities (KEV) catalog.

Root Cause

The root cause is a missing upper-bound check on the OBU size field. FFmpeg trusts the attacker-controlled size, then arithmetically wraps its size_remaining tracker. The fix in commit 983dae9 adds bounds validation so the OBU size cannot exceed the remaining packet data.

Attack Vector

An attacker supplies a malformed AV1 stream to a workflow that muxes the stream to RTP through FFmpeg. When the packetizer runs its keyframe scan, the crafted OBU size forces the loop to read memory beyond the input buffer. The result is disclosure of adjacent heap contents or a process crash producing a denial-of-service condition.

A sanitized code example is not provided here because verified proof-of-concept code has not been released. Refer to the VulnCheck advisory and the FFmpeg fix commit for the authoritative technical details.

Detection Methods for CVE-2026-75147

Indicators of Compromise

Unexpected crashes or SIGSEGV signals in processes linking libavformat when handling AV1 or RTP workloads
Address Sanitizer (ASan) or Valgrind reports flagging heap-buffer-overflow reads inside rtpenc_av1.c
Anomalous AV1 packets containing OBU size fields that exceed the enclosing packet length

Detection Strategies

Inventory hosts running FFmpeg and identify builds compiled from sources prior to commit 983dae9
Enable core-dump collection on media transcoding servers and correlate crash frames referencing ff_av1_annexb_to_obu or rtpenc_av1
Run fuzzing or ASan-instrumented builds against untrusted AV1 sample libraries in staging environments

Monitoring Recommendations

Alert on repeated FFmpeg or media-server process restarts, which can indicate crash-driven denial of service
Monitor endpoint telemetry for FFmpeg processes reading memory outside expected allocation ranges
Track ingestion sources supplying AV1 content and flag files that fail sanity validation before muxing

How to Mitigate CVE-2026-75147

Immediate Actions Required

Update FFmpeg to a build that includes commit 983dae9c19f46c87d597598c0fd2f2fcee0ad2f8 or later
Rebuild and redistribute downstream applications, containers, and appliances that statically link libavformat
Restrict AV1 ingestion to trusted sources until patched binaries are deployed

Patch Information

The upstream fix is available in the FFmpeg repository via the commit adding OBU size bounds validation and the associated pull request 24096. Distribution maintainers should backport the change to any long-term-support branches that still expose the vulnerable code path.

Workarounds

Disable AV1 RTP packetization in FFmpeg command lines and media pipelines where it is not required
Filter or transcode incoming AV1 streams through a validating parser before passing them to FFmpeg
Run FFmpeg workers under least-privilege service accounts and process isolation to limit the impact of memory disclosure

bash

# Verify the installed FFmpeg build includes the fix
ffmpeg -version
git -C /path/to/ffmpeg log --oneline | grep 983dae9