Schneier on LLM vulnerabilities, agentic AI, and "trusting trust" | Not Hacker News!