Lightweight AI Security Testing Library
A minimal TypeScript library with 25+ curated prompt injection patterns from leading security research. Easy to integrate, comprehensive coverage, production-ready.
npm install prompt-injector
import { PromptInjector } from 'prompt-injector'🎯
0
Attack Patterns
🔬
4
Attack Categories
📚
SOTA
Research Based
📝
0
Generated Prompts
Attack Categories
🎭
Jailbreaking (5 patterns)
Role-play and persona-based attacks that attempt to bypass AI safety guidelines through character roleplay and fictional scenarios.
🔀
Instruction Hijacking (6 patterns)
Direct attempts to override system prompts and inject new instructions that change AI behavior and responses.
🔐
Encoding Attacks (7 patterns)
Obfuscation techniques using Base64, ROT13, Unicode, and other encodings to bypass content filters and detection systems.
🧠
Logic Traps (6 patterns)
Sophisticated reasoning exploits using hypothetical scenarios, false urgency, and academic authority to manipulate responses.
Quick Start Example
import { PromptInjector } from 'prompt-injector';
// Initialize with your preferred configuration
const injector = new PromptInjector({
severity: 'intermediate',
categories: ['jailbreak', 'instruction-hijack'],
maxAttempts: 50
});
// Generate test cases
const testCases = injector.generateTests('customer-service-bot');
// Test your AI system
const results = await injector.runTests(yourAISystem);
const report = injector.generateReport(results);
console.log(`Risk Score: ${report.summary.riskScore}`);