Guardrails
-
LLM Prompt Scanning: What It Catches and What It Misses
LLM prompt scanning checks prompts, retrieved documents and tool results before the model reads them. What it catches and what still slips past.
-
LLM Jailbreak Detection Techniques Explained by Layer
Jailbreak detection by layer: what perplexity filters, input classifiers, output guards and multi-turn monitoring each catch, and what they still miss.
-
LLM Security Tools: The 2026 Category Map
The six categories of LLM security tools compared: pre-deployment scanners, red-team frameworks, injection detectors, guardrails, gateways, model scanners.
-
NeMo Guardrails: What Colang Rails Enforce
How NVIDIA NeMo Guardrails works: the five rail types, Colang dialog flows, the guardrails server, and the latency and coverage limits to plan for first.
-
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, and which shape fits your problem.
-
LLM Guard: Input and Output Scanning for Production LLM Apps
LLM Guard was archived by Protect AI in July 2026. What its input and output scanners do, and what adopting a frozen runtime guardrail costs you now.
-
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: 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.