Security & Risk

Hallucination, injection, leakage, red teaming — the model-level risk map

• Hallucination — Why models confabulate by construction, the shapes it takes, what actually reduces it — and how to measure it instead of hoping. (4 lessons, 45 min)
• The Single-Channel Problem — The root vulnerability: one token stream with no type system separating instructions from data — and why training alone cannot close it. (3 lessons, 40 min)
• Prompt Injection and Jailbreaks: The Model-Level View — Direct and indirect injection, the jailbreak taxonomy, and why refusals are trained dispositions rather than enforcement. (4 lessons, 45 min)
• Data Leakage and Memorization — Models memorize training text and can be made to emit it; prompts flow to providers; deletion is hard. The leakage map, both directions. (3 lessons, 40 min)
• The OWASP Top 10 for LLM Applications — The community-standard risk list for LLM applications, walked entry by entry and mapped onto this site — plus how to actually use it. (3 lessons, 45 min)
• Red Teaming LLMs — Adversarial probing as a discipline: manual and automated methods, coverage, and turning findings into fixes. (3 lessons, 40 min)
• In Production: The Defense Stack Around a Served Model — The domain’s production capstone: defense in depth around a model that cannot defend itself — provider filters, app-side controls, and monitoring. (3 lessons, 40 min)