✓What system prompt leakage is — in plain English and in OWASP’s definition
✓What developers embed in system prompts and why it creates risk
✓Why “keep it secret” is not a security control — and the right mental model to replace it
✓LLM07 vs LLM02 — the distinct mechanism and why the fix differs
✓The Bing Chat “Sydney” incident (Feb 2023) — direct extraction in a single turn
✓GitHub Copilot prompt extraction (May 2023) — multi-step probing from a patient attacker
✓Encoding-based extraction — how Base64 and roleplay frames bypass naive output filters
✓EchoLeak (CVE-2025-32711) — zero-click chained injection exfiltrating full context via email
✓All 3 OWASP attack scenarios — credentials, security bypass, competitive intelligence
✓All 6 mitigations — what OWASP says, which real incidents each would have stopped, how to validate