# AI Security & Safety

> AI security and safety tooling: red-teaming frameworks like Garak and PyRIT, guardrails like NeMo Guardrails and Lakera Guard, privacy tooling like Microsoft Presidio, and the standards practitioners build against, from the EU AI Act to the OWASP Top 10 for LLMs.

21 tools in 4 subcategories. Web page: https://ailandscape.org/category/ai-security-safety · Each tool also has a markdown version at /tool/{slug}.md

## Offensive Testing & Red Teaming

- [ART](https://ailandscape.org/tool/art): Adversarial Robustness Toolbox by IBM
- [CleverHans](https://ailandscape.org/tool/cleverhans): Library for benchmarking adversarial ML attacks
- [Foolbox](https://ailandscape.org/tool/foolbox): Fast adversarial attacks for deep neural networks
- [Garak](https://ailandscape.org/tool/garak): LLM vulnerability scanner and red-teaming framework
- [HarmBench](https://ailandscape.org/tool/harmbench): Standardized evaluation framework for LLM robustness
- [PyRIT](https://ailandscape.org/tool/pyrit): Python Risk Identification Toolkit for GenAI by Microsoft
- [TextAttack](https://ailandscape.org/tool/textattack): NLP adversarial attacks, augmentation, and training

## Guardrails & Defense

- [Guardrails AI](https://ailandscape.org/tool/guardrails-ai): Framework for adding guardrails to LLM outputs
- [Lakera Guard](https://ailandscape.org/tool/lakera-guard): Real-time prompt injection and jailbreak protection
- [LlamaGuard](https://ailandscape.org/tool/llamaguard): Meta's LLM-based input/output safeguard
- [NeMo Guardrails](https://ailandscape.org/tool/nemo-guardrails): NVIDIA's toolkit for adding safety guardrails to LLMs
- [Patronus AI](https://ailandscape.org/tool/patronus-ai): Enterprise LLM evaluation and safety platform
- [Prompt Shield](https://ailandscape.org/tool/prompt-shield): Microsoft Azure's prompt injection detection service
- [Shieldstral](https://ailandscape.org/tool/shieldstral): Mistral's open-weights multimodal safety classifier for text and images

## Privacy & Compliance

- [Microsoft Presidio](https://ailandscape.org/tool/microsoft-presidio): PII detection and anonymization for text and images
- [OpenDP](https://ailandscape.org/tool/opendp): Differential privacy library for data analysis
- [PySyft](https://ailandscape.org/tool/pysyft): Federated learning and privacy-preserving ML
- [TensorFlow Privacy](https://ailandscape.org/tool/tensorflow-privacy): Training ML models with differential privacy

## Standards & Frameworks

- [EU AI Act](https://ailandscape.org/tool/eu-ai-act): EU's comprehensive regulatory framework for AI
- [NIST AI RMF](https://ailandscape.org/tool/nist-ai-rmf): NIST AI Risk Management Framework
- [OWASP Top 10 for LLMs](https://ailandscape.org/tool/owasp-top-10-for-llms): Top 10 security risks for LLM applications

---

Part of [AI Landscape](https://ailandscape.org), an open map of the AI ecosystem. Index for AI assistants: https://ailandscape.org/llms.txt