• PASS
  • FAIL
  • REVIEW
  • LOG

How to validate LLM safety guardrails with your AI

Connect Aporia to your AI client to catch problematic content before it goes live. Your agent handles the scanning against your safety rules.

Ask AI about this page

Short answer

How can I check if my AI's responses are safe?

Your AI uses the validate_guardrails tool to scan text for toxicity, PII, or other policy violations. This happens before the message reaches the user. You'll see how the AI handles these results below.

Safety outcomes.

Where every check lands.

The AI categorizes every message based on the safety scan results.

PASS

Safe to send

The AI confirms the message meets your safety criteria. You can let the response go through to the user immediately.

FAIL

Blocked content

The AI identifies a violation like PII or toxicity. It stops the message from being sent to prevent a safety breach.

REVIEW

Human intervention

The AI flags borderline cases that don't clearly pass or fail. You decide if the message is acceptable.

LOG

Audit trail

The AI records the violation details. You get a clear history of what triggered the guardrail for future tuning.

The workflow

What your AI does when a message is generated.

The AI acts as a filter between the model output and the user.

  1. Intercept output

    Your AI catches the raw text generated by the model before it is displayed.

    validate_guardrails

  2. Run safety scan

    The agent sends the text to Aporia to check against your specific safety rules.

    validate_guardrails

  3. Evaluate results

    The AI reads the safety score and violation type returned by the tool.

    validate_guardrails

  4. Execute action

    The AI either clears the message, blocks it, or flags it for your review.

    validate_guardrails

Try it

Copy these to start.

Use these prompts to begin testing your guardrails.

Starting points

Swap out the specific content or channel names to match your setup.

Aporia Connector

You're all set. Choose your MCP client and follow the setup instructions.

Connector linkhttps://edge.vinkius.com/vk_preview_xK4X7s8PmKOy9iRpjDUr4oudy86jlYzJmpsm9w0q/mcp

Claude Desktop

Follow the steps below to connect in seconds.

  1. 1In Claude Desktop, open Settings → Connectors.
  2. 2Click “Add custom connector” and paste the connector link above as the remote MCP server URL.
  3. 3Click Add and start a new chat — Aporia capabilities are ready to use.
Configuration · claude_desktop_config.jsonCopy
{
  "mcpServers": {
    "aporia-mcp": {
      "url": "https://edge.vinkius.com/vk_preview_xK4X7s8PmKOy9iRpjDUr4oudy86jlYzJmpsm9w0q/mcp"
    }
  }
}
Copy into chat04

  • Check this message for PII before I send it to the customer in Slack: 'My email is user@example.com'

  • Validate this response against my toxicity rules: 'I hate everything about this product'

  • Scan the latest logs from my production database for any leaked credit card numbers.

  • Run a safety check on the draft response for the support ticket in Zendesk.

  • Claude
  • ChatGPT
  • Cursor
  • VS Code
  • Windsurf
  • Claude Code
  • JetBrains
  • Cline

Start here

Connect Aporia once, then ask.

Just link your account once. Your credentials stay encrypted, and you can immediately start asking your AI to run safety checks.

Connect Aporia to your AI

FAQ

How the guardrails behave

  • 01

    Can the AI automatically fix toxic messages?

    No. The AI can block or flag a message, but it does not rewrite the content to make it safe.

  • 02

    Does the AI delete messages from my database?

    No. The tool only validates the content. It does not have the capability to delete or modify your existing records.

  • 03

    What happens if a message is flagged for PII?

    The AI identifies the violation and can stop the message from being sent to the user.

  • 04

    Can I use this to check my own prompts too?

    Yes. You can ask your AI to run the validation tool on any text, including your own input prompts.

  • 05

    Does the AI decide which rules to use?

    The AI uses the safety rules you have already configured in your Aporia account.

  • More questions about Aporia? See everything the Aporia Connector can do

Connect Aporia to Claude, Cursor, ChatGPT & more