Why it matters

Runtime protection catches issues at execution. Understanding shapes defense-in-depth.

Advertisement

The architecture

Input scanners: prompt injection, secrets, malformed.

Output scanners: PII, harmful content, prompt leak.

LLM Guard flowInput scannersprompt injection + secretsLLM callif input cleanOutput scannersPII + harmfulSimilar tools: NVIDIA NeMo Guardrails, Meta Purple Llama
LLM Guard architecture.
Advertisement

How it works end to end

Scanners: modular; enable per use case.

Alternatives: NeMo Guardrails, Purple Llama.

Performance: extra latency per scan.

False positive tuning needed.