Developing Genuine Security Instincts in LLMs

Experts analyze why current AI safety fails and how to move from brittle rule-following to intuitive, human-like threat detection in LLMs.