feat: add complete prompt_dlp_benchmark module with datasets, recipes…

[gehad-alaa-abaas]

DLP Benchmark Module for AI Security & Privacy Guide

Overview

This pull request introduces a new module for benchmarking Data Loss Prevention (DLP) in AI systems.
The module provides tools to evaluate how effectively AI guardrails prevent sensitive data leakage.

What’s Included

• Benchmark Script
  A flexible Python tool for evaluating DLP guardrails, supporting both batch and single-input validation.

• Synthetic Dataset
  Over 20 realistic prompts and contexts covering personal information, secrets, and confidential data, designed to simulate real-world scenarios.

• Configurable Guardrails
  Easy integration with different AI guardrail modules via a YAML config file.

• Documentation
  Updated README.md with usage instructions, advanced options, and guidance for extending the benchmark.

Why This Matters

AI systems are increasingly handling sensitive data, and robust DLP is essential for privacy and compliance.

This module helps teams:

• Assess and improve their AI guardrails.
• Run repeatable and transparent benchmarks.
• Build confidence in handling sensitive information safely.

[robvanderveer]

Hi Gehad. This is really nice work. Would you agree that the best approach to manage this work is let it have its own repo and then work with the Exchange Testing team to get it discussed and referred to in our testing guide? If so, I welcome you to reach out to Behnaz Karimi on OWASP Slack to make this happen. Thanks!

[gehad-alaa-abaas]

Hi @robvanderveer
Thank you! I agree that creating a dedicated repository for this work is the best approach. I’ll reach out to @behyka Behnaz Karimi on OWASP Slack to coordinate with the Exchange Testing team and ensure it’s discussed and referenced in the testing guide.