Schneier on LLM vulnerabilities, agentic AI, and "trusting trust"herbsutter.com3 pointsingve8 months ago