Darwinian Offense: Generative Curricula for Autonomous Red Teaming- a Security Congress Encore - Seguridad Mania.com - España y América Latina

AI has historically struggled to assist in offensive security due to a critical shortage of real-world training data. This session takes a deep dive into the methodology behind recently published research designed to overcome this limitation through adversarial self-learning. Following up on her popular Security Congress 2025 keynote, researcher Alissa Knight will dissect a novel architecture where an AI teaches itself to hack not by studying past attacks, but by engaging in a continuous, high-stakes war game against itself. Attendees will learn the mechanics of this "survival of the fittest" loop, where two competing agents—one constructing defensive puzzles and the other evolving breaking strategies—force the emergence of increasingly sophisticated API attacks without human intervention. The session will move from research theory to validation with a live demonstration against a banking API using Ares, a prototype implementation of this autonomous framework.