← All talks

Fuzzing Power: AI Boosts Security Testing & Exploit Triage #shorts

BSides Frankfurt1:241.1K viewsPublished 2025-12Watch on YouTube ↗
About this talk
Fuzzing finds clear, verifiable errors like buffer overflows, giving strong security signals. AI can optimize search strategy, understand semantics, and aid exploitability triage. Offense can train with RL models to succeed. #Fuzzing #Cybersecurity #AISecurity #Exploitability #BufferOverflow #cybersecurity #coding #ai #machinelearning
Show transcript [en]

They love fuzzing because I think this is where you get really strong signals at least in some of the areas. So, you know, easy buff overflow. you've got the asen every time there's a crash you get this this error and it's a very clear mechanical verifier something is wrong on your stack and you know I think AI can geni can help you know if you remember this graph in the search space to have the right search strategy not just to um you know to spray and pray for instance but understand the semantics of this application based on semantics understand oh do I need now do I need now if this is a um you asky um string

which is accepted do I do I need now to try out with unic code instead of just doing pulling some random strings inside it can help as well with exploitability triage which is becoming a bit more difficult right bypassing ISLR and so on it's not very easy so this is where it's going to to become very hard to verify but I think it's a good start and just mapping down this complexity of all these verifiers looking at all this offense capabilities we see that most of them they're hard, but most of them are still possible to generate. They are very binary whether you have achieved something or you haven't achieved something. And that's really important

based on this area because it means offense can very easily train up based on RL the models to succeed.