EPISODE 27: Ignore all previous instructions: Threat modelling AI Systems

Welcome to the grand finale of season two of ‘Compromising Positions’, where we delve into the fascinating world of AI security. In this special episode, your hosts will guide you through the labyrinth of securing AI models, one step at a time. For those who prefer a quick overview, we offer an abridged version on Apple Podcasts and Spotify. This version deep dives into two key topics: Jeff’s unique mnemonic C-PTSD for threat modeling AI systems, and an intriguing discussion on the correlation between boredom, worm-killing, and AI efficiency gains. For those who crave a deeper dive, scroll down or visit our Youtube channel for the extended cut. This version includes everything from the regular version, plus: Jeff’s academic journey in AI at the University of Hull Lianne’s preparation for a 100 days of Code in Python for her MSc in Data Science and AI at Leeds Trinity University A critical discussion on OpenAI’s transparency and the latest AI wearable technology, along with the complexities of consent and privacy in an ‘always recorded’ lifestyle Whether you choose the regular or extended version, we appreciate your support throughout season two. Stay tuned for more enlightening discussions in season three! Thank you for being a fantastic audience.