retr0regHeap Overflow in FFmpeg EXIF
> Pwno is a AI cybersecurity startup...
We all know that LLMs were used to find these vulnerabilities, specifically on high impact projects. That's fine.
However, my only question is who actually provided the patch: The maintainers of FFmpeg? The LLM that is being used? Or the security researchers themselves after finding the issue?
It seems that these two statements about the issue are in conflict:
> We found and patched 6 memory vulnerabilities in FFmpeg in two days.
> Dec, 2025: avcodec/exif maintainer provided patch.
> We all know that LLMs were used to find these vulnerabilities
How do we know that? You seem quite certain.
They pitch their company as finding bugs "with AI". It's not hard to point one of the coding agents at a repo URL and have it find bugs even in code that's been in the wild for a long time, looking at their list that looks likely to be what they're doing.
The list is pretty short though for 8 months. ossfuzz has found a lot more even with the fuzzers often not covering a lot of the code base.
Manually paying people to write fuzzers by hand would yield a lot more and be less expensive than data centers and burning money, but who wants to pay people in 2026?
Bugs are not equivalently findable and different techniques surface different bugs. The direct comparison you're trying to draw here doesn't hold.
It does not matter what purported categories buffer overflows are in when manual fuzzing finds 100 and "AI" finds 5.
If Google gave open source projects $100,000 per year for a competent QA person, it would cost less than this "AI" money straw fire and produce better results. Maybe the QA person would also find the 5 "AI" detected bugs.
This would make sense if every memory corruption vulnerability was equivalently exploitable, which is of course not true. I think you'll find Google does in fact fuzz ffmpeg, though.
Google gives a pittance even for full ossfuzz integration. Which is why many projects just have the bare minimum fuzz tests. My original point was that even with these bare minimum tests ossfuzz has found way more than "AI" has.
Another weird assumption you've got here is that fuzzing outcomes scale linearly with funding, which, no. Further, the field of factory-scale fuzzing and triage is one Google security engineers basically invented, so it's especially odd to hold Google out as a bad actor here.
At any rate, Google didn't employ "AI" to find this vulnerability, and Google fuzzing probably wouldn't have outcompeted these researchers for this particular bug (totally different methods of bugfinding), so it's really hard to find a coherent point you'd be making about "fuzzers", "AI", and "Google" here.
My guess is the main "AI" contribution here is to automate some of the work around the actual fuzzing. Setting up the test environment and harness, reading the code + commit history + published vulns for similar projects, identifying likely trouble spots, gathering seed data, writing scripts to generate more seed data reaching the identified trouble spots, adding instrumentation to the target to detect conditions ASan etc don't, writing PoC code, writing draft patches... That's a lot of labor and the coding agents can do a mediocre job of all of it for the cost of compute.
https://x.com/FFmpeg/status/2006773495066464580
> Seeing as this has made the orange site, let it be known this person is a model security researcher.
> The issue was not in any FFmpeg release, and a report was sent three days after a new code was added to FFmpeg Git.
> There was no big CVE ADVISORY "MUH SECURITEH" "you need to fix this now or you will be hacked and the world will end" associated with the report.
FFmpeg (@FFmpeg) on X
Seeing as this has made the orange site, let it be known this person is a model security researcher. The issue was not in any FFmpeg release, and a report was sent three days after a new code was added to FFmpeg Git. There was no big CVE ADVISORY "MUH SECURITEH" "you need to
This is another drawback of security research, but one that had already existed before "AI" with ossfuzz.
You basically cannot commit in public to the main branch and audit and test everything 3 months before a release, because any error can be picked up, will be publicized and go into the official statistics.
> ... go into the official statistics.
There are no "official" statistics. None of this matters. If we judged projects by the number of security holes they had, then no one would be using ffmpeg, which had hundreds of serious vulns.
Vulnerability research is useful insofar that the bad guys are using the same techniques (e.g., the same fuzzing tools), so any bugs you squash make it harder for others to attack you. If your enemy is a nation state, they might still pack your laptop / phone / pager with explosives, but the bar for that is higher than popping your phone with a 0-day.
Vulnerability research is demonstrably not useful for improving the security of the ecosystem in the long haul. That's where sandboxing, hardening, and good engineering hygiene come into play. If you're writing a browser or a video decoder in C/C++, you're going to have exploitable bugs.
> Vulnerability research is demonstrably not useful for improving the security of the ecosystem in the long haul. That's where sandboxing, hardening, and good engineering hygiene come into play. If you're writing a browser or a video decoder in C/C++, you're going to have exploitable bugs.
IMHO, vulnerability research is the stick that drives the ecosystem towards all those things. Reports of vulnerabilities in the codec for Rebel Assult videos (or whatever) leads one to disable codecs other than those they need. Reports of vulnerabilities in playlist support leads one to disable playlist support where it's unnecessary and run transcodes in a chroot sandbox with no network access. Reports of buffer oveflows leads one to prefer implementation in memory safe languages where available with sufficient performance and also to sandbox when possible.