AI Security & Guardrails

Protect AI systems and the data that flows into them: guardrails, usage control, red-teaming and model scanning.

Up to five options per category, in alphabetical order: the order means nothing and nothing here is a ranking, a recommendation or an endorsement. We earn nothing from any vendor. For wider market coverage, see analyst reports such as Gartner Magic Quadrants and Forrester Waves, and test any tool against your own needs.

AI guardrails and prompt-injection protection

Filters what goes into and out of AI models to block prompt injection and unsafe output.

Name	What it is	Type
Lakera	Runtime protection against prompt injection for AI applications.	Paid
Meta Llama Guard and Prompt Guard	Open-source safety classifiers for model input and output.	Open source
NVIDIA NeMo Guardrails	Open-source toolkit for adding rules to LLM applications.	Open source
Pillar Security	Security platform for AI applications and agents.	Paid
Prisma AIRS (Palo Alto Networks)	Palo Alto Networks' runtime security for AI applications.	Paid

AI data protection and AI usage control (shadow AI, DLP for AI)

Stops sensitive data going into AI tools and shows which AI tools staff use.

Name	What it is	Type
Harmonic Security	Sees and controls employee use of AI tools and the data shared with them.	Paid
LLM Guard	Open-source scanners for prompts and model output.	Open source
Microsoft Presidio	Open-source detection and anonymisation of personal data in text.	Open source
Nightfall	Data loss prevention for SaaS and AI tools.	Paid
Prompt Security	Controls for employee AI use and AI applications.	Paid

AI red-teaming and model scanning

Probes AI models and applications for weaknesses before attackers do.

Name	What it is	Type
garak	Open-source scanner for LLM vulnerabilities.	Open source
HiddenLayer	Model scanning and AI red-teaming.	Paid
Mindgard	Automated red-teaming for AI systems.	Paid
promptfoo	Open-source tool for testing and red-teaming prompts and models.	Open source