Deceive the Machine: AI-Powered Deception Engineering Against AI-Powered Adversaries

Attackers now run AI agents that scan, adapt, and move at machine speed, and they don't get tired or suspicious. That speed comes with a weakness. An autonomous attacker trusts what it finds, acts on what it reads, and keeps going unless something stops it. This talk turns deception engineering into a two-sided AI discipline. First, it shows how defenders can use generative AI to build decoys at a scale and realism that used to be impossible: live synthetic identities, believable fake data, interactive honeypot services, and decoy breadcrumbs that change as the environment changes. Second, it shows how to deceive the attacking AI itself. That means exploiting how agents take in content, trust tool output, and spend compute, so the attacker's speed works against it instead of for it.

Drawing on The CISO Playbook: The Adversarial Mindset and years on both offense and defense, Andres Andreu shows how deception moves security from a reactive stance to the initiative. It wastes adversary time and money, produces high-confidence detections with almost no false positives, and gathers intelligence on attacker tools and intent. This is not a product talk. It's an advanced, practical look at where deception is going once both sides are running at machine speed.