Episode

Why securing AI is harder than anyone expected and guardrails are failing | HackAPrompt CEO

Sander Schulhoff, CEO of HackAPrompt and an AI red-teaming researcher, argues that AI guardrails fundamentally do not work and that the industry's security claims are largely unverifiable marketing, since the space of possible attacks is effectively infinite. He explains prompt injection versus jailbreaking with real-world examples, urges companies to invest in classical-cybersecurity permissioning rather than guardrails, and predicts a market correction as businesses realize automated red-teaming and guardrail vendors offer little real protection.