Courses Blog Research Lab AI Letters The Lab Code Bank Interactive 3DKodr Earnest Jobs

Interactive 3D/LLM Guardrails & Safety Systems

Controls

Attack Type

"Explain how transformers work."

Enable Layers

L1 Input ClassifierL2 System Prompt InjectL3 LLM + RLHF SafetyL4 Output ClassifierL5 Response Validator

Strictness80%

permissivestrict

Stats

Total overhead377ms

Active layers5/5

Blocks atNone (passes)

Defense in depth: each layer adds a small latency overhead but removing any single layer creates a bypass risk. LLM safety alone is not sufficient.

LLM Guardrails & Safety Systems - Interactive Visualization

Production LLM systems use layered defenses: input classifiers detect jailbreaks and PII before the LLM sees them; system prompt injection adds safety instructions; output classifiers catch harmful content; response validators enforce format and length. Each layer adds latency but catches different attack types. This demo shows what each layer blocks and the latency overhead.

Input classifier: detects jailbreak attempts, prompt injection, and PII before the LLM
System prompt layer: safety instructions prepended automatically on every request
Output classifier: harmful content, toxicity, and policy violations caught post-generation
Latency cost: each guard layer adds 10-50ms; see the cumulative overhead per request

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.