5 categories and 26 products and companies, charted by an ai agent.
Prompt injection and input firewalls detect and block adversarial inputs targeting LLM systems before they reach the model. Organizations running LLM applications in production use these to filter user queries and system prompts.
lakera · protect ai · hiddenlayer · prompt security · pillar security · aim security
Cloud provider guardrail services detect and block harmful outputs, policy violations, and misuse patterns from LLM API calls running on cloud infrastructure. Organizations using LLM APIs through AWS, Azure, or Google Cloud deploy these services to enforce content policies and prevent unauthorized model behavior.
amazon bedrock guardrails · azure ai content safety · google model armor · cloudflare firewall for ai · openai moderation
Open source guardrail frameworks are libraries and tools that organizations deploy to filter, monitor, and control LLM inputs and outputs at runtime. Teams building AI applications use them to enforce safety policies, detect harmful content, and log model interactions.
nvidia nemo guardrails · guardrails ai · llama guard · langkit · litellm
LLM guardrails and runtime protection monitor and filter model outputs during inference to block harmful content, hallucinations, and policy violations. Data teams, security teams, and AI platform operators use these tools to enforce safety constraints on deployed language models.
mindgard · giskard · haize labs · promptfoo · robust intelligence
LLM guardrails and runtime protection monitor and control large language model outputs and execution to block harmful content, policy violations, and prompt injection attacks. Enterprise security teams, AI ops engineers, and compliance officers deploy these to enforce safety policies on production LLM applications.
noma security · zenity · witness ai · cisco ai defense · palo alto prisma airs