Mar 19, 2026 · 10:15 PM · 100 Queens Quay East, Toronto
LLMs Don't Hack, They Guess
Jeet — lessons from building offensive AI for multi-layered codebases
RSVP for this eventSeats remaining shown after RSVP launches.
Speaker
Jeet
Offensive Security Engineer at Robinhood. Started in network engineering, moved to offensive security as a consulting pentester. Current focus: where AI and LLMs actually help (and don't) in offensive…
About
LLMs Don't Hack, They Guess
AI tools can find the obvious bugs. They follow the most suspicious-looking signal until it runs out. This talk looks at what it actually means to use LLMs and agents for vulnerability discovery in complex, multi-layered codebases — what works, what doesn't, and why.
Where AI genuinely shines: pattern-based vulnerability detection, surfacing suspicious code, accelerating early triage during reviews. Where it hits hard limits: anything that requires exhaustive reasoning across inputs, logic, and execution paths. That's where mixed findings and noise explode — AI finds a vulnerable sink and thinks it found a 0-day crit.
Practical strategies for blending human intuition with AI to actually find bugs, plus a clear-eyed take on why current agent designs structurally miss the long-tail vulnerabilities.
This event has concluded.