Rebuff
Prompt injection detection API for LLM applications
Verdict
Purpose-built for prompt injection detection — a real attack vector in RAG systems. Uses a canary token technique alongside an LLM classifier. Early-stage; combine with input sanitization.
Other Guardrails & Safety
- Anthropic Constitutional AIStable
Principle-based self-improvement for harmlessness and helpfulness
- Azure AI Content SafetyProduction
Microsoft's content moderation API for text and images
- Guardrails AIStable
Add input/output validation and safety rails to LLM calls
- Lakera GuardStable
Real-time prompt injection and jailbreak protection API
- Llama Guard 3Experimental
Meta's fine-tuned safety classifier for prompt and response screening
- Microsoft PresidioProduction
Data protection and anonymization for PII in LLM pipelines