Prompt injection happens when text controlled by an attacker is treated by the model as instructions instead of data. Models cannot reliably separate the two, so it sits behind many public agent exploits, and no known filter fully prevents it. After reading you should be able to explain why and name the defenses that actually reduce risk.
Read in this order
- OfficialLLM01:2025 Prompt InjectionStart here: the standard definition, attack types, and mitigations in one page.
- ArticleComment and Control: Prompt injection to credential theft in Claude Code, Gemini CLI and GitHub Copilot Agent (Aonan Guan)One pull request comment, three coding agents, stolen credentials: the April 2026 case to start with.
- OfficialOWASP LLM Prompt Injection Prevention Cheat SheetA checklist of defensive measures you can apply to an app today.