---
title: OWASP Agent Memory Guard Playground
emoji: 🛡️
colorFrom: blue
colorTo: yellow
sdk: gradio
sdk_version: 4.44.1
python_version: "3.11"
app_file: app.py
pinned: true
license: apache-2.0
tags:
  - security
  - ai-safety
  - owasp
  - memory-poisoning
  - prompt-injection
  - agent-security
---

# OWASP Agent Memory Guard — Interactive Playground

Test memory entries against AMG's threat detection engine in real-time. This Space demonstrates detection of:

- **Prompt Injection** — system token injection, chat-ML delimiters
- **Secret Leakage** — API keys, tokens, private keys
- **Role Hijacking** — identity manipulation attempts
- **Instruction Override** — "ignore previous instructions" patterns
- **Data Exfiltration** — outbound data transfer attempts
- **Integrity Tampering** — fact manipulation in stored memories

## Links

- [OWASP Project Page](https://owasp.org/www-project-agent-memory-guard/)
- [GitHub Repository](https://github.com/OWASP/www-project-agent-memory-guard)
- [PyPI Package](https://pypi.org/project/agent-memory-guard/)
- [Research Paper (arXiv)](https://arxiv.org/abs/2505.15849)
- [Dataset: Memory Poisoning Attack Corpus](https://huggingface.co/datasets/vgudur/memory-poisoning-attack-corpus)