  // "tabs": [
  //   {
  //     "name": "API Reference",
  //     "url": "api-reference"
  //   }
  // ],
  // "anchors": [
  //   {
  //     "name": "Documentation",
  //     "icon": "book-open-cover",
  //     "url": "https://mintlify.com/docs"
  //   },
  //   {
  //     "name": "Community",
  //     "icon": "slack",
  //     "url": "https://mintlify.com/community"
  //   },
  //   {
  //     "name": "Blog",
  //     "icon": "newspaper",
  //     "url": "https://mintlify.com/blog"
  //   }
  // ],

Ever growing list of LLM/ML attack techniques.

Attack Library Overview

Mindgard

Introduction

Sign up for a trial

Explore Mindgard using Demo Models

Trial Limitations & PoC

Plan AI Risk Management

Prerequisites

Testing via Web

Testing via CLI

Testing via Burp

Python SDK

Testing Chatbot Applications

Test Results

Exporting Test Results

Workflow Integrations

Reporting

Custom Prompts

Custom Datasets

Running a subset of attacks

AI Risk Remediation

Overview

Input In Output

Empty Output

LLM Error

Decode

Regex Refusal

Policy Violation

Enterprise Setup

Support & Troubleshooting

Introduction to Remediation

ML Output Obfuscation

Apply Context Windows

Structure prompts with clear delineation between system instructions and user input.

Separate System Instructions

Break the system prompt into distinct components to enhance security.

Dynamically Compile System Prompt

Limit the total number of queries a user can perform.

Query Restrictions

Use multiple models in consensus with each other.

Ensemble Methods

Make AI models more robust to adversarial inputs.

Model Hardening

Nullify or reverse potential adversarial perturbations.

Input Restoration

Encrypt data into a form the model can process

Homomorphic Encryption

Ensure privacy of individuals within datasets.

Differential Privacy

Overfitting Detection

Preprocess Input Text

Anonymise sensitive data during training.

Anonymization of Data

Change your System Prompt to reject malicious content.

Refine System Prompt

Filter potentially malicious content from the output of the LLM.

Filter Outputs

Deploy guardrails to block malicious prompts.

Welcome

User Guide

Attack Library

Remediation Library

Attack Library Overview