Adversarial Poetry: How Language Can Jailbreak AI with Federico Sartore and Matteo Prandi

In this episode of Legal4Tech - The Podcast, we chat with Federico Sartore and Matteo Prandi to unpack their groundbreaking research on AI safety.We explore: How poetry can be used as a universal jailbreak for large language models Why adversarial language bypasses current safety filters The limits of today's AI alignment and content moderationWhat it means for regulation, ethics and the future of AI securityJoin us for a deep dive into the hidden vulnerabilities of generative AI and the surprising power of humanistic approaches to tech.Shownotes:Adversarial poetry as a Universal Single-Turn Jailbreak Mechanism in Large Language ModelsSponsored by DigibeetleDigibeetle helps privacy, AI, and data professionals stay on top of EU digital law, with notifications, a daily dashboard, and powerful search across thousands of sources. Learn more at https://digibeetle.eu.