Critical FFmpeg zero-click RCE vulnerabilities found by autonomous security agent
Security firm Depthfirst discovered 21 zero-day vulnerabilities in FFmpeg, the widely used media processing library, including a critical RCE-capable heap buffer overflow in its AV1 RTP depacketizer. These flaws can be exploited with a single 183-byte RTP packet and impact media ingest pipelines, CCTV, and cloud transcoding services that process untrusted RTSP or RTP streams. Patches are advised for system administrators.
Key Takeaways
- One high-severity RCE flaw (DFVULN-127) can be triggered with no user interaction or authentication via a malformed RTSP stream.
- Eight of the 21 vulnerabilities have been assigned CVEs, including flaws in the TS demuxer, VP9 decoder, and DASH demuxer.
- Multiple bugs discovered by Depthfirst's autonomous agent had remained latent in the FFmpeg codebase for over 20 years.
- The security scan cost approximately $1,000 in compute, a 90% cost reduction compared to previous AI-driven audits of the library.
- Systems processing untrusted remote video feeds, such as surveillance and cloud ingest services, are advised to apply patches immediately.
Why It Matters
FFmpeg is arguably the most critical piece of invisible infrastructure in the streaming ecosystem, powering everything from browser-based playback to enterprise transcoding. This mass disclosure of RCE-capable flaws highlights a significant shift in the threat landscape as autonomous AI agents begin to weaponize and automate the discovery of deep-seated architectural bugs in mature C/C++ codebases. For streaming providers, this necessitates a more aggressive posture toward patching low-level libraries and potentially isolating ingest pipelines that handle unauthenticated user streams. Watch for a rise in similar AI-generated vulnerability reports as the economics of automated security auditing continue to collapse.
Additional Context
The detection of these 21 vulnerabilities follows a sustained effort by major tech players to apply advanced machine learning to open-source security. In early 2026, Google’s Big Sleep team disclosed 13 bugs in FFmpeg, while Anthropic’s Mythos model successfully identified a 16-year-old vulnerability in the library’s H.264 implementation, per Anthropic's May 2026 report. These discoveries underscore the vulnerability of the roughly 1.5 million lines of heavily optimized C code that comprise FFmpeg, which has traditionally relied on manual audits and standard fuzzing for protection. This spike in disclosures arrives amid a record-breaking period for software patching. In June 2026, Google released Chrome 149, which addressed 429 vulnerabilities—the highest number in a single release for the browser. Per Forbes (May 2026), this trend is largely driven by a flood of AI-generated security reports that have forced companies like Google to overhaul their bug bounty programs to prioritize concise, reproducible proofs-of-concept over lengthy automated write-ups. The discovery by Depthfirst also highlights the economic disparity in modern cybersecurity. While Anthropic reportedly spent $10,000 on its initial FFmpeg audit using Mythos, specialized agents can now perform comparable threat modeling for approximately $1,000. This drop in costs suggests that even small teams or malicious actors could soon conduct exhaustive scans of critical infrastructure. Security observers note that while defenders are using these tools to find flaws first, the speed of automated discovery is rapidly outpacing the ability of volunteer maintainers to develop and ship patches for high-impact open-source projects.
Read full article at cybersecuritynews.com
Get this in your inbox → Subscribe
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source