---
title: AgentGuard Bench
emoji: 🛡️
colorFrom: blue
colorTo: indigo
sdk: gradio
sdk_version: 4.36.0
app_file: app.py
pinned: false
license: apache-2.0
tags:
  - llm-security
  - prompt-injection
  - guardrails
  - agentic-ai
  - safety
---

# 🛡️ AgentGuard Bench

A side-by-side comparison of a **pattern-based** prompt-injection
guardrail and a **fine-tuned transformer classifier**, built to make
the latency/robustness tradeoff between the two approaches concrete
and explorable.

- **Live Compare** — paste any prompt, see both guardrails score it instantly
- **Adversarial Test Suite** — run a curated set of attack + benign prompts and compare accuracy and latency
- **About** — methodology, related work, and limitations

The pattern-based guardrail is a Python port of the `PromptInjectionGuardrail`
I contributed to [LangChain4j](https://github.com/langchain4j/langchain4j)
(Java) in [PR #5619](https://github.com/langchain4j/langchain4j/pull/5619) —
same detection taxonomy, demonstrating it generalizes across ecosystems.
The model-based guardrail uses
[`protectai/deberta-v3-base-prompt-injection-v2`](https://huggingface.co/protectai/deberta-v3-base-prompt-injection-v2)
(Apache 2.0).

See the **About** tab inside the Space for full methodology, citations,
and known limitations.