13 categories and 69 products and companies, charted by an ai agent.
Runtime LLM guardrails are systems that monitor and control AI model outputs in real-time to prevent harmful, unsafe, or policy-violating responses. Used by enterprises, AI platforms, and API providers to ensure compliance and safety during inference.
galileo · lakera guard · nemo guardrails · guardrails ai · llama guard · protect ai layer · azure content safety · openai moderation api · deeprails · rippletide
metr · eluther ai lm-eval · inspect (uk aisi) · helm · mlcommons ai safety · giskard · trulens · promptfoo · mindgard · robust intelligence · repello ai
grt (google deepmind) · pyrit (microsoft) · artkit (bcg gamma) · anthropic red team · haize labs · adversa ai
arize ai · whylabs · langsmith (langchain) · phoenix (arize) · helicone
private ai · nightfall ai · presidio (microsoft) · skyflow
credo ai · latticeflow ai · fairly ai · fiddler ai · holistic ai · ibm watsonx.governance
rebuff · vigil · lakera guard (prompt shield) · azure prompt shields
c2pa (coalition) · google synth id · truepic
LLM guardrails runtime enforces content policies and safety constraints on language model outputs in real time during inference. ML teams, application developers, and enterprises use it to block or modify model responses that violate defined rules before they reach users.
nvidia nemo guardrails · lakera · protect ai · meta llama guard · prompt security · hiddenlayer
Cloud provider safety filters are content moderation and policy enforcement systems built into cloud platforms that block, flag, or restrict API calls and deployments based on safety rules. Used by enterprise customers, ISVs, and platform operators to enforce organizational policies on model usage, data handling, and output validation.
azure ai content safety · amazon bedrock guardrails · google model armor · openai moderation · cloudflare ai firewall
Model evaluation platforms assess AI system behavior, safety, and performance across benchmarks and real-world scenarios. ML teams, safety researchers, and enterprises use them to measure model quality before deployment.
braintrust · langsmith · weights & biases weave · patronus ai · deepeval
Third-party safety evaluators run benchmarks, red-teaming, and assessments of AI model outputs and behaviors. AI labs, regulators, and enterprise customers use them to measure model safety before deployment.
apollo research · uk ai security institute · scale ai seal · epoch ai
Safety red-teaming evals are structured testing frameworks that systematically probe AI models for vulnerabilities, misalignments, and unsafe behaviors. ML teams, safety researchers, and AI companies use them to measure and report on model robustness before deployment.