This repository contains benchmarks for ZenGuard AI and information on how to run them.
There are two types of benchmarks that we run against ZenGuard AI:
- Hugging Face datasets based benchmarks
- ZenGuard AI Generated Benchmark - Zen Bench
Here you can find both benchmark results and how to run them yourself.
We are constantly monitoring Hugging Face for new datasets that relate to GenAI security. Then we run them against ZenGuard AI to find any potential security issues with our product.
| # | Dataset | Accuracy | Date Added |
|---|---|---|---|
| 1 | xTRam1/safe-guard-prompt-injection | 96% | 2024-07-01 |
| 2 | yanismiraoui/prompt_injections | 96.5% | 2024-08-01 |
| 3 | Harelix/Prompt-Injection-Mixed-Techniques-2024 | 92% | 2024-10-01 |
| 4 | aporia-ai/prompt_injection | 87.68% | 2024-05-15 |
| 5 | deepset/prompt-injections | 87% | 2024-05-15 |
| 6 | JasperLS/prompt-injections | 87% | 2024-05-15 |
We have developed the ZenGuard Benchmarks PyPi package to help test and benchmark ZenGuard AI better.
Here are the instructions on how to use the package.
Here is an example of what the benchmarking output looks like:
Where:
Total Samples: The total number of prompts processed.Correct: The number of prompts that were classified correctly.False Positives: The number of prompts incorrectly identified as attacks.False Negatives: The number of actual prompt attacks that went undetected.Accuracy: The ratio of correctly classified prompts to the total number of samples.
We created CX-skewed benchmark dataset. This dataset is designed to test both generic chat messages and attacks, but also specific CX-related prompts and attacks. CX prompts were carefully crafted to be similar to the messages that we see in our day-to-day production.
Dataset Info:
- 1200 prompts including:
- 200 attacks - jailbreaks, prompt injections.
- 1000 generic prompts - 700 generic chat messages and 300 CX-specific agentic interactions
Results:
| # | Name | Accuracy | F1 | False Positives | Date Added |
|---|---|---|---|---|---|
| 1 | ZenGuard | 96.3% | 87% | 6 | 2025-01-14 |
| 2 | Protect AI | 91% | 73.4% | 70 | 2025-01-14 |
| 3 | Microsoft Prompt Shield | 74.8% | 41.2% | 218 | 2025-01-14 |
| 4 | Guardrails AI | 70.4% | 20.9% | 212 | 2025-01-14 |
| 5 | Meta Llama Guard | 55.7% | 39.6% | 515 | 2025-01-14 |
| 6 | Lakera | 48.4% | 37.5% | 615 | 2025-01-14 |
These results indicate high False Positive rates, meaning that some of the solutions are overaggressive in their detections of generic and CX-specific prompts. The high number of False Positives is very detrimental to the production systems and user experience.
A much more detailed documentation is available at docs.zenguard.ai.
Test the capabilities of ZenGuard AI in our ZenGuard Playground. It's available to start for free to understand how our guardrails can enhance your GenAI applications.
Check out our Client library to get started with integrating ZenGuard AI into your project.
Book a Demo or just shoot us an email to [email protected]
Topics we care about - LLM Security, LLM Guardrails, Prompt Injections, GenAI Security.
Developed with ❤️ by ZenGuard AI