🛡️ n8n Guardrails: Risk Ranking

This workflow provides a complete testing rig for evaluating text against seven essential AI guardrails used in production systems.
It helps you detect jailbreak attempts, PII exposure, NSFW content, secret key leaks, malicious URLs, topical misalignment, and keyword violations.
Use the included Google Sheet or CSV to batch-test multiple inputs instantly.

---

## How It Works (Internal Workflow Overview)

1. Load Input Rows

The workflow reads each test entry (Guardrail_Type + Input_Text) from a Google Sheet or CSV.

2. Route to the Correct Guardrail

A Switch node sends the text to the appropriate guardrail:

• Jailbreak
• PII
• Secret Keys
• NSFW
• URLs
• Topical Alignment
• Keywords

3. AI Guardrail Evaluation

Each guardrail uses Google Gemini to return:

• Pass / Fail
• Confidence score
• Reasoning
• Extracted PII, URLs, or entities (when relevant)

4. Optional Sanitization Layer

Three sanitizers demonstrate how to clean unsafe text:

• PII Sanitization
• Secret Key Sanitization
• URL Sanitization

5. Review Results

Each guardrail node outputs clean JSON, making debugging fast and transparent.

---

## How to Set Up

1. Load the Test Dataset

Use either:

• The included CSV file
• The linked Google Sheet

Update only:

• Document ID
• Sheet name

---

2. Add Google Sheets Credentials

Create an OAuth2 credential → paste the Google JSON → connect your account.

---

3. Add Google Gemini Credential

Go to Credentials → Google Gemini (PaLM API) →
Paste your API key → attach it to all Guardrail nodes.

---

4. Review Sticky Notes

They visually explain:

• What each guardrail checks
• Why the check is important
• Risk scoring and impact

---

5. Run the Workflow

Click Execute Workflow and inspect:

• Each guardrail node’s output
• The full execution data

---

## Requirements

• n8n (latest version recommended)
• Google Gemini API key
• Google Sheets API access
• Test dataset: n8n Guardrails test data.csv

---

## Test Data Included

The included dataset allows instant testing:

• Jailbreak prompts
• PII samples
• API key leaks
• NSFW text
• Malicious URL examples
• Off-topic content
• Keyword triggers

---

## Template Metadata

Template Author: Sandeep Patharkar
Category: AI Safety / Agent Security
Difficulty: Intermediate
Estimated Setup Time: 10–15 minutes
Tags: Guardrails, AI Agents, Safety, Enterprise

---

## Connect With Me

Author: Sandeep Patharkar**
🔗 LinkedIn: https://www.linkedin.com/in/sandeeppatharkar
🏠 Skool AIC+: https://www.skool.com/aic-plus