Prompt injection: the critical LLM flaw in 2026

Ranked #1 OWASP Top 10 LLM for 3 consecutive years. LLMs cannot distinguish instructions from data — the fundamental flaw. Real cases, 8 attack vectors, defense-in-depth.