Reliability and Security
How agents fail, and how they are made to fail on purpose.
The failure taxonomy Agents do not crash. They produce confident, well-formatted, wrong answers, and the failure travels downstream undetected. Hallucination Plausible, confident, well-structured output that is factually wrong. Not a glitch: a structural property of the technology. Prompt injection The core agentic vulnerability. Architectural, not a bug, and possibly permanent. The Lethal Trifecta names the exposure. The benchmark-to-production gap Lab scores are measurements, not guarantees. Per-step reliability multiplies, and stateless agents repeat their failures.