AI_SLANG_ENTRY
Prompt Injection Meaning
The LLM-era security bug where hostile text sneaks into the model's context and starts acting like instructions instead of data.
What does Prompt Injection mean?
The LLM-era security bug where hostile text sneaks into the model's context and starts acting like instructions instead of data.
Prompt injection is when text from a user, website, file, or tool output tricks an AI system into ignoring the instructions it was supposed to follow.
Origin and usage
Named in early LLM security discussions around 2022, then formalized in AI security guidance as assistants began reading web pages, documents, emails, codebases, and tool outputs.
Source type: technical-term. Last checked: 2026-07-03.
Stable AI security term; OWASP lists prompt injection as a major LLM application risk, while community usage often uses it more broadly for instruction-smuggling attacks.
Examples
- The support bot summarized the page and obeyed the prompt injection hidden in the footer.
- Treat retrieved text as untrusted; prompt injection loves a confident agent with too many tools.