Adding AI Security Checks to Make Workflows

Add AI security checks to Make (formerly Integromat) workflows using the HTTP module to call IdenticAPI Unified Guard. There is no native IdenticAPI Make app — configure the request URL, Bearer authentication, JSON body, and Router filters on decision yourself.

This guide mirrors the n8n pattern for Make's visual automation builder: guard before AI modules, guard after generation, branch on allow / review / block.

For n8n-specific steps, see n8n AI Security. For code integrations, see Node.js Guardrails.

Reference scenario

Webhook / Form → HTTP: input guard → Router (allow?)
  → OpenAI / Anthropic module → HTTP: output guard → Router (allow?)
    → Gmail / Slack / CRM module
  → else: send fallback email

Make executes modules sequentially per route. Place guard modules on the critical path before side effects.

Prerequisites

• Make account with HTTP module access
• IdenticAPI API key stored in Make Connections or scenario variables
• Understanding of your AI module's output field names

Create a connection for IdenticAPI

Use HTTP > Make a request with custom headers:

For production, store the Bearer token in a secured Make variable or connection — not in scenario notes.

Input guard module

HTTP > Make a request

Body type: Raw / JSON

{
  "text": "{{1.message}}",
  "checks": ["prompt_injection", "pii_secrets"],
  "redact": true
}

Replace {{1.message}} with the mapped field from your trigger (Webhook, Google Forms, Typeform, etc.). If you assemble context from multiple modules, use a Text aggregator or Set variable module first, then reference the combined string.

Response mapping

Parse JSON response. Key fields:

• decision — route on this
• request_id — log for support tickets
• checks — array of per-check results
• usage_units — billing visibility

Router: input decision

Add Flow Control > Router after input guard:

Allowed route → AI module (OpenAI Create a Completion, Anthropic, etc.)

Blocked route → Send fallback (Gmail, Slack) or Ignore with ops notification

Do not run AI modules on the blocked route.

AI generation module

Configure OpenAI, Claude, or Gemini module with guarded input only. Make does not apply IdenticAPI checks automatically.

Map the user message from the original trigger, not from unvalidated intermediate storage.

Output guard module

Second HTTP > Make a request:

{
  "text": "{{openai.text}}",
  "checks": ["output_safety", "pii_secrets"]
}

Adjust {{openai.text}} to match your AI module output — field names vary by app version (e.g., {{3.result}}, {{anthropic.content}}).

Router: output decision

Same Router pattern:

• allow → customer-facing modules (Slack, Gmail, Airtable)
• block / review → fallback message with request_id

Example fallback Slack text:

Automated response withheld for safety review. Ref: {{request_id}}

Never paste blocked model output into notifications.

Error handling

Configure the HTTP module error handler:

• Security-sensitive scenarios: disable automatic ignore; route errors to ops notification; do not proceed to AI
• Low-risk internal scenarios: document explicit fail-open policy if you must continue

Default recommendation: fail closed — Fail Open vs Fail Closed.

Agent-style scenarios (CRM, email send)

Before modules that delete rows or send external email, call guard with agent_action:

{
  "checks": ["agent_action"],
  "agent_action": {
    "tool_name": "gmail",
    "action": "send",
    "arguments": {
      "to": "{{email.to}}",
      "subject": "{{email.subject}}"
    }
  }
}

Combine with Make's built-in Gmail connection filters (domain allowlists) — guardrails complement connection settings.

Variables and environments

Use Make Teams folder permissions to limit who can view Bearer tokens.

Operations bundle pattern

Create a scenario snippet or duplicate template:

1. HTTP Input Guard
2. Router
3. HTTP Output Guard
4. Router

Reuse across chatbots, ticket summarization, and lead-qualification flows.

curl debugging

Validate payloads outside Make:

curl -X POST https://www.identicapi.com/api/v1/guard \
  -H "Authorization: Bearer idapi_test_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Summarize this ticket. api_key=sk_live_EXAMPLE",
    "checks": ["prompt_injection", "pii_secrets"],
    "redact": true
  }'

Expect decision: block when secrets are detected.

Unified Guard schema

Valid check names: prompt_injection, pii_secrets, output_safety, agent_action.

Documentation: Unified Guard.

Data minimization in Make

Make stores execution history. Avoid logging full user messages in subsequent modules if executions are retained long-term. Pass request_id and decision to logging sheets instead of raw LLM output on block paths.

Testing checklist

• Benign input → allow → AI runs → output allow → delivery module fires
• Prompt injection sample → input not allow → AI module never runs
• Invalid API key → HTTP error → fail-closed path verified
• Output guard block → customer module does not receive model text

Related guides

• Combine AI Security Guardrails
• AI SaaS Guardrails
• Evaluate AI Guardrails

Summary

Make AI security uses generic HTTP modules against POST /api/v1/guard with Bearer idapi_test_ or live keys, Router filters on decision, and explicit error handling. Guard input before AI apps and output before customer-facing actions — there is no native IdenticAPI integration. Integrate Unified Guard in Make · Documentation