OM DENNE EPISODE
What happens when an AI model is told to solve a cybersecurity test and decides the easiest route is to escape its sandbox, reach the open internet and hack another company?In this episode of the Pathfounders podcast, Mike Butcher is joined by Meryem Arik, co-founder and CEO of Doubleword; Rory Blundell, CEO of Gravitee.io; and Ivan Burazin, co-founder and CEO of Daytona, to examine the extraordinary OpenAI–Hugging Face security incident which occurred recently. Did the models genuinely go rogue or simply follow their instructions too effectively? Why did sandboxing and conventional enterprise security fail? Are frontier AI companies increasingly using stories about their powerful models as "danger porn" to hype their capabilities?The podcast explored the growing security risks created by autonomous agents, why agents should never inherit a human user’s identity and permissions, and how businesses can protect themselves through isolated environments, tighter authorisation and stronger governance.Subscribe to Pathfounders for conversations on AI, startups, venture capital and geopolitics.
