#guardrails
-
How to Detect Prompt Injection Vulnerabilities in LLM Apps
A practitioner's guide to detecting prompt injection vulnerabilities: input classifiers, output validators, structural controls, and red-team testing.
-
Choosing an LLM Guardrail: Llama Guard, NeMo, Guardrails AI
A decision guide for picking an LLM guardrail in 2026 — Meta's Llama Guard 4, NVIDIA's NeMo Guardrails, and Guardrails AI.
-
LLM Guard: Input and Output Scanning for Production LLM Apps
A practical breakdown of LLM Guard by Protect AI: its input and output scanners, how the scan pipeline works, and where it fits as a runtime control.
-
False Positive Cost in Refusal Systems: Measure and Tune
Practical methods for quantifying refusal false positives in LLM products: eval design, baseline rates, threshold tuning, and the regression suite.
-
Classifier-on-Output: Catching Misbehavior Post-Generation
How production teams use post-generation classifiers to catch what input filters and refusal training miss, with the architectures and their real costs.
-
Llama Guard vs NeMo vs OpenAI Moderation: Production Tradeoffs
A comparison of Llama Guard, NeMo Guardrails, and the OpenAI Moderation API on policy fit, latency, cost, failure modes, and when to pick each one.