Add LLM Guardrails

[rafaelsandroni]

Feature Request

API based options:

• https://lakera.AI
• https://guardion.ai
• https://guardrailsai.com

Open source models from HuggingFace:

• https://huggingface.co/protectai/deberta-v3-base-prompt-injection-v2
• https://github.com/meta-llama/PurpleLlama/tree/main/Prompt-Guard

Motivation

Some specific use cases require security controls, whether its a RAG agent or an AI agent. Typically, I conduct red teaming and evals to ensure my product is secure, but absolute certainty is never guaranteed. The emerging pattern now is to implement guardrails and specialized SLMs to prevent prompt attacks at the input level and data leakage at the output.

Your Contribution

No response

[rogeriochaves]

hey @rafaelsandroni, we recently included LangWatch (https://langwatch.ai/) as a Langflow components:

#4722

it has prompt injection detection, PII and other guardrails as well as evaluators

also, here is a prompt template using the loop component to run an evaluation through a set of example, could be useful for red teaming too:

#6761