AI agent discovers 21 zero-days in FFmpeg for $1,000 as Chrome patches record 429 security bugs

AI Agent Uncovers Decades-Old Software Vulnerabilities in FFmpeg

A security startup called depthfirst deployed an autonomous AI agent that discovered 21 previously unknown zero-days in FFmpeg, the open-source media library embedded in nearly every application that processes video 1

1

. The AI agent scanned FFmpeg's roughly 1.5 million lines of C code and produced reproducible proof-of-concept inputs for each vulnerability, all for approximately $1,000 in compute costs 2

2

. Several of these software vulnerabilities had been latent in the codebase for 15 to 20 years, with one stack overflow in the service-description-table code dating back to 2003—sitting untouched for 23 years 1

1

.

Most of the AI-discovered bugs are heap or stack overflows in parsers and demuxers, spanning components from the TS demuxer to the VP9 decoder 2

2

. Nine vulnerabilities already carry CVE identifiers, numbered CVE-2026-39210 through CVE-2026-39218, while the remaining bugs have been fixed upstream but await formal numbering 1

1

. Depthfirst has published proof-of-concept code demonstrating each flaw, underscoring the practical threat these vulnerabilities pose to systems ingesting untrusted media streams.

Chrome Ships Record-Breaking Patch as AI-Generated Reports Flood In

The same week brought another milestone in vulnerability discovery: Google shipped Chrome 149 with patches for 429 security bugs, the highest count ever in a single browser release 1

1

. Over 100 of these are critical or high severity, predominantly use-after-free and insufficient input validation flaws 1

1

. The most severe, CVE-2026-10881 with a CVSS score of 9.6, is an out-of-bounds read and write in the ANGLE graphics engine that allows a crafted page to escape Chrome's sandbox and execute code on the host system 2

2

. Google paid $97,000 for this critical report 1

1

.

While Google hasn't directly attributed the 429 vulnerabilities to AI, the company overhauled its bounty program in April after experiencing a flood of AI-generated reports 1

1

. The revised program now requests concise reproducers instead of the lengthy writeups that AI systems typically generate 2

2

. Of the 22 critical bugs patched, 19 were found internally by Google, while only 10 of roughly 90 high-severity bugs came from external researchers 1

1

.

Autonomous AI Agents Reshape Cybersecurity Landscape

FFmpeg is not new to AI in identifying software flaws. Google's Big Sleep agent reported multiple FFmpeg bugs last year, now visible on the project's security page tagged BIGSLEEP 1

1

. Anthropic's Mythos model extracted a 16-year-old H.264 flaw and other vulnerabilities from FFmpeg for about $10,000, three of which shipped in FFmpeg 8.1 1

1

. Depthfirst claims to have achieved comparable results at a tenth of that cost 2

2

.

The pattern extends beyond FFmpeg and Google Chrome. Days ago, another autonomous tool identified an authenticated remote code execution flaw in Redis that had remained undetected since version 7.2.0, hiding for over two years 1

1

. A February study demonstrated that an AI agent could reproduce working proof-of-concepts for more than half of 100 real Linux kernel N-day bugs, outperforming traditional fuzzing techniques 1

1

. Mozilla recently patched 271 Firefox vulnerabilities found by Mythos in a single pass 2

2

.

The Growing Gap Between Discovery and Remediation

The core challenge has shifted dramatically. AI uncovers zero-day vulnerabilities at unprecedented speed and minimal cost, but triaging AI-generated reports, shipping fixes, and deploying patches remains slow and resource-intensive 2

2

. Much of this work still depends on volunteers and a thin layer of human triagers now expected to keep pace with machines 1

1

. For FFmpeg, users should pull the fixed upstream build or distribution security updates immediately, prioritizing systems that ingest untrusted RTSP or AV1-over-RTP streams 1

1

. FFmpeg is widely bundled in media pipelines, Python wheels, container images, and appliances, making embedded copies a critical concern beyond system packages 1

1

.

For Chrome users, updating to version 149.0.7827.53 on Linux or 149.0.7827.53/54 on Windows and macOS is essential, or confirming that auto-update has executed 1

1

. The response must match the new pace: shorter patch cycles, auto-update wherever possible, and treating dependency bumps that carry CVE fixes as security work rather than routine maintenance 1

1

. The question is no longer whether AI can find the bugs, but whether security teams can fix them fast enough 2

2

.