HackerNicholas Carlini - Black-hat LLMs | [un]prompted 2026
unpromptedYouTube ↗
Further reading 1
Caption notes
Carlini presents language-model-assisted vulnerability discovery, including SQL injection and a kernel heap overflow, and discusses what these cases mean for defenders.
Original passages · 7
you can do it just by like asking the model to find these bugs.it will probably find the same bug each time.I can't report because I haven't validated them yet.if I put a safeguard in place that's very very weak,in the limit I'll like you know formally verify all of my protocols.find and exploit zero-day vulnerabilities in very important pieces of softwarecurrent models are better vulnerability researchers than I am