AI Agent Finds 21 Zero-Day Flaws in FFmpeg Media Library

In June 2026, an autonomous AI security agent developed by startup depthfirst identified 21 confirmed zero-day vulnerabilities in the FFmpeg media library ^[1]. The agent scanned through 1.5 million lines of C code to locate these critical flaws ^[1]. This breakthrough illustrates the rapidly growing role of artificial intelligence in detecting software vulnerabilities within massive codebases ^[2].

The discovery is part of a broader shift in cybersecurity where AI agents automate complex research tasks ^[2]. These tools are now being used to define the weekly landscape of threats and defenses ^[2]. Industry experts note that such automated discoveries are becoming a standard feature in modern information security product cycles ^[3].