# AI Security

> Keeping AI safe from people who try to break it on purpose.

**Field:** Artificial Intelligence  
**Category:** Safety & Security  
**Also known as:** Security for AI, LLM Security, AI Application Security

## Think of it like

It is like the locks on a house. They keep out people who want to break in.

## Definition

AI security is keeping AI systems safe from people who attack them on purpose. Attackers may trick a model with hidden words. They may poison its data, steal the model, or use an AI agent's tools to reach your files. AI safety asks if the AI itself does harm. AI security asks if someone else can make it do harm. Both are needed, and they use different tools.

## Example

Lucy's team builds a support chatbot. Before launch, they test if a customer can make it show other people's orders or use tools it should not.

## Why it matters

An AI linked to data and tools is a new door into a company. Attackers look for it.

## Common mistake

Thinking the usual app checks cover it. A model can be talked into things a normal program never could. So it needs its own tests.

## Three principles

- **Attacks on purpose.** People trying, not accidents.
- **Words as weapons.** Text alone can be an attack.
- **Old rules still count.** Least access, logs and checks.

## Related terms

- [AI Safety](https://learn.benibix.com/term/ai/ai-safety)
- [Prompt Injection](https://learn.benibix.com/term/ai/prompt-injection)
- [Data Poisoning](https://learn.benibix.com/term/ai/data-poisoning)
- [Model Extraction](https://learn.benibix.com/term/ai/model-extraction)
- [OWASP Top 10 for LLM Applications](https://learn.benibix.com/term/ai/owasp-top-10-for-llm-applications)
- [Principle of Least Privilege](https://learn.benibix.com/term/development/principle-of-least-privilege)

## Told apart from

- [AI Safety vs AI Security](https://learn.benibix.com/compare/ai-safety-vs-ai-security): People think AI safety and AI security are two words for the same job. They are not. Safety asks if the AI itself could cause harm, through a wrong goal or a mistake. Security asks if an attacker could make it cause harm, with a trick or stolen access. A model can be well meaning and still easy to attack, so a team needs both.

![Two panels. On the left, AI safety: the AI causes harm by itself, through a wrong goal or a mistake. On the right, highlighted, AI security: an attacker outside sends a trick to the AI and makes it cause harm.](https://learn.benibix.com/diagrams/ai4-safety-vs-security.svg)

---

Source: https://learn.benibix.com/term/ai/ai-security
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-18
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
