# Agentic AI Protection

Wallarm provides API-first security for AI systems by protecting AI agents, AI proxies, and APIs with AI features by preventing injection attacks and data leakage, controlling costs, and ensuring secure, compliant operations.

![Agentic AI in work - schema](https://docs.wallarm.com/images/agentic-ai-protection/agentic-ai-schema.png)

## Common attacks on AI Agents

Common attacks on AI Agents include:

* Jailbreaks:

    * Retrieval of hidden system prompts and instructions for exploitation.
    * Injection of encrypted prompt commands to bypass content filters.
    * Invocation of restricted APIs by an agent for unauthorized operations.

* Attacks on Agent APIs:

    * Attacks and exploits tools used by agents using common API attacks.
    * Sensitive data leaks through internal APIs.
    * Weak authentication and misconfiguration exploitation.

* Bots and Agent Abuse:

    * Automated bot attacks including low-and-slow attacks and DDoS.
    * Usage abuse and credits overages, including license abuse.
    * Automated account takeover attacks.
    * Mass prompt injection.

* Rogues and shadow AI Agents:

    * Agents deployed by shadow IT lack proper security hardening, leaving backdoors for attackers.
    * Cross-tenant data leaks by unauthorized agents in shared environments.
    * Exploitation of unprotected shadow agents risks credit theft and massive infra bills.

See detailed description of Wallarm's Agentic AI Protection of the official site [here](https://www.wallarm.com/solutions/s-protect-agentic-ai).

## How protection works

Wallarm's protection against attacks on AI Agents works in a few simple steps:

1. You deploy Wallarm [filtering node](https://docs.wallarm.com/about-wallarm/overview.md#how-wallarm-works) using the [appropriate deployment option](https://docs.wallarm.com/installation/supported-deployment-options.md).
1. Optionally, you enable [automatic discovery](https://docs.wallarm.com/agentic-ai/agentic-ai-discovery.md) of AI/LLM endpoints in your API inventory by enabling and Wallarm's [API Discovery](https://docs.wallarm.com/api-discovery/overview.md).
1. In Wallarm Console, you create one or several [AI payload inspection](https://docs.wallarm.com/agentic-ai/ai-payload-inspection.md) mitigation controls defining how to detect [AI-agent attacks](https://docs.wallarm.com/attacks-vulns-list.md#attack-types) and mitigate them.
1. For MCP servers, you configure [MCP mitigation controls](https://docs.wallarm.com/agentic-ai/mcp-mitigation-controls.md) to enforce access policies, validate request parameters, and ensure tool calls conform to the published schema.
1. Wallarm automatically detects attacks and [performs action](https://docs.wallarm.com/agentic-ai/ai-payload-inspection.md#mitigation-mode) (just register an attack or perform blocking by IP or session).
1. Detected and blocked attacks are displayed in [API Sessions](https://docs.wallarm.com/api-sessions/overview.md). In the malicious request details, the back-link to the policy that caused detection and/or blocking is presented.

![API Sessions - session with detected malicious AI payload](https://docs.wallarm.com/images/agentic-ai/api-sessions-system-prompt-retrieval.png)

## Demo

[Explore the Agentic AI attack mitigation demo →](https://rsa-demo-playground.darkmatter.wallarm.tools/)

![Wallarm against attacks on Agentic AI - demo](https://docs.wallarm.com/images/agentic-ai-protection/agentic-ai-wallarm-demo.png)

This demo demonstrates the following cases:

* Exploit BOLA through the agent
* System prompt retrieval and business logic abuse
* Identity impersonation and tool misuse

On completing any of the scenarios, Wallarm detects the attack and mitigates them in the correspondence with the mitigation mode - you obtain a link to the [**API Sessions**](https://docs.wallarm.com/api-sessions/overview.md) section of Wallarm Playground, where you can explore the session and the attack inside it.