There's a piece of software called FFmpeg that you've almost certainly never heard of, but almost certainly use every day. It's an open-source media library that handles the grunt work of encoding, decoding, and processing video and audio. It's inside VLC, Chrome, Firefox, Android, macOS, countless streaming apps, and probably your smart TV firmware. It is, in the most literal sense, everywhere.

This week, a security startup ran an autonomous AI agent against FFmpeg's codebase and it came back with 21 previously unknown vulnerabilities — all zero-days, meaning no patch existed at the time of discovery. That's not a record-setting finding in a niche obscure package. That's a significant haul in critical shared infrastructure that underpins a huge proportion of the software stack modern life runs on.

What the AI agent actually did is the interesting part. Rather than fuzzing inputs until the program crashes — the traditional approach — the agent appears to have combined static code analysis with dynamic testing, reasoning about the code the way a human researcher would, but faster and without coffee breaks. The result was 21 bugs found in a compressed timeframe that would have taken a human team substantially longer.

The same week, Google shipped Chrome 149 with patches for 429 security vulnerabilities — the most ever fixed in a single Chrome release. To be clear, the two events aren't directly linked; the Chrome bugs weren't found by the same AI. But the proximity is hard to ignore. Both data points point to the same underlying reality: the attack surface of widely-deployed software is far larger than our existing security processes have been able to see.

Why this matters beyond the headline number. FFmpeg vulnerabilities are particularly dangerous because they sit at the media parsing layer — which means attackers can often trigger them by getting a target to open a malicious video file. No phishing link, no credential theft required. Just: watch this video. That attack pattern has been used in real-world exploitation before; the library's ubiquity makes each new bug in it a potential multi-platform problem.

The Chrome patch count is staggering for different reasons. 429 bugs in a single release suggests either a massive backlog of deferred fixes, or a step-change in the rate at which Google's internal security teams (and external researchers) are finding flaws. Given that AI-assisted fuzzing has been part of Google's Project Zero toolkit for years, the latter is plausible.

The bigger signal here is about the economics of vulnerability research changing. If an AI agent can find 21 zero-days in a week, the cost of vulnerability discovery is dropping — for defenders, yes, but also for anyone with bad intentions and access to similar tooling. The defenders found these first. That won't always be the case.

For Australian organisations: FFmpeg is deeply embedded in broadcast infrastructure, streaming platforms, and government AV systems. The ACSC's guidance on software supply chain risk is directly applicable here — if you're running software that bundles FFmpeg, watch for vendor advisories closely over the coming weeks as patches roll out.