Exposing LLM Vulnerabilities

What if the very tools designed to make us smarter are also making us vulnerable? A single prompt. A subtle tweak. A forgotten language. That’s all it takes. As LLMs weave themselves into the fabric of our daily lives, their promise feels limitless — until you look beneath the surface. In this wide-ranging and quietly urgent conversation, Kate and Andrew explore the evolving landscape of AI vulnerabilities, from adversarial attacks and prompt injections to multilingual blind spots and poisoned training data. They share stories from real-world projects, reflect on the role of collaborative tools in catching threats early, and unpack why even small teams must prioritise security from day one. Together, they don’t just highlight what can go wrong — they illuminate the pathways forward. This is a thoughtful, human-centred episode about risk, responsibility, and the power of working together in a rapidly changing world. This is one of those episodes that stays with you long after the headlines fade.