# What is Prompt Injection? Definition & Technical Mechanics

> **Source:** [SPIDITS AI AI Glossary](https://spidits.com/ai-glossary)  
> **Category:** Model Limitations  
> **Canonical Citation:** [https://spidits.com/ai-glossary/prompt-injection](https://spidits.com/ai-glossary/prompt-injection)

## Definition
Prompt Injection is a security vulnerability where a malicious user provides input that overrides the pre-configured system instructions or safety alignment filters of a Large Language Model, hijacking its control flow.

## Architectural Mechanics & Deep Dive
Prompt injection is a security vulnerability where a user inserts malicious text instructions into a prompt to override the LLM's system guardrails and safety directives. This can cause the model to bypass safety alignment, generate restricted content, execute unauthorized function calls, or exfiltrate session data to external endpoints.

## Industry Applications & Real-World Use Cases
Vulnerability analysis, LLM firewall design, and input sanitation.

## Frequently Asked Questions

### What is the difference between direct and indirect prompt injection?
Direct prompt injection occurs when a user prompts the model to ignore rules. Indirect prompt injection happens when a model reads untrusted external data (like a webpage) containing malicious instructions.

### How do you defend against prompt injection?
By separating user input from developer instructions, using strict system prompt isolation, and employing secondary filter models to inspect outputs.

## Related Terminology
- [adversarial-attack](https://spidits.com/ai-glossary/adversarial-attack)
- [guardrails](https://spidits.com/ai-glossary/guardrails)
- [ai-safety](https://spidits.com/ai-glossary/ai-safety)

---
*From the SPIDITS AI Knowledge Base (336+ terms indexed). Explore full technical definitions: [https://spidits.com/ai-glossary](https://spidits.com/ai-glossary)*