Akto LLM Guard

LLM-Guard is a comprehensive tool designed to fortify the security of Large Language Models (LLMs).

Features

• Prompt Injection Detection: Detect and prevent prompt injection attacks
• PII Detection: Identify and handle personally identifiable information
• Toxicity Detection: Filter harmful and toxic content
• Code Detection: Identify and manage code in prompts
• Data Leakage Prevention: Prevent sensitive information leakage
• URL Scanning: Detect malicious URLs in outputs
• Bias Detection: Identify biased content in model outputs

Installation

pip install akto-llm-guard

For ONNX runtime support:

pip install akto-llm-guard[onnxruntime]

For GPU support:

pip install akto-llm-guard[onnxruntime-gpu]

Quick Start

from llm_guard.input_scanners import PromptInjection

scanner = PromptInjection()
sanitized_prompt, is_valid, risk_score = scanner.scan("Your prompt here")

if is_valid:
    print("Prompt is safe")
else:
    print(f"Prompt failed security check. Risk score: {risk_score}")

Repository

GitHub: https://github.com/akto-api-security/llm-guard

License

MIT License