Automating security work: from hand-written rules to LLM agents

Abstract

This lecture follows how security work has been automated, from hand-written rules and programs, to machine learning, to large language model (LLM) agents. Part 1 traces the three eras and where each one stopped. Part 2 explains how LLM agents work and why making them better at security is still hard. Part 3 covers two ways to make a model better at this work: the harness of tools and prompts around the model, and the training of the model itself.

Date
Nov 18, 2026
Event
Guest lecture, CMPSC 279 F26
Location
University of California, Santa Barbara

The slides open in the browser. Use the arrow keys or the space bar to move through them, press F for full screen and ? to see all the keys.

Saastha Vasan
Saastha Vasan
AI Security Researcher

My research interests include malware analysis, machine learning, threat intelligence and program analysis.