Snehal Antani, Horizon3 | Crowdstrike Fal.Con 2026
summary
Argues that security must move from point-in-time checks to a continuous, adversary-informed loop where AI-driven offensive agents probe production systems and automated blue-team remediation hack-fix-verify repeatedly; research shows frontier offensive models interact with decoys up to 92% decoy interaction versus 37% for humans and that optimally placed deception reaches a 95% catch rate; digital twins miss messy live edge cases; Base64 interview trick demonstrates LLM manipulation.
AI-written summary · about
