Moderate User Content with the OpenAI Moderation API

The OpenAI Moderation API is a free endpoint that classifies text across eleven harm categories including hate speech, harassment, violence, sexual content, and self-harm. Running user input through the moderation endpoint before passing it to your chat completion is a defense-in-depth measure that catches obviously harmful content without consuming expensive GPT tokens. This example shows a moderation request and explains how to interpret the category scores and flagged boolean. The moderation response returns a flagged boolean and per-category scores from 0.0 to 1.0. A score above the threshold for any category sets flagged to true. The threshold for each category is tuned by OpenAI for precision/recall balance, but you can implement your own thresholds using the raw scores — for example, flagging content at a lower threshold for children's applications or at a higher threshold for adult platforms that allow mature content. For most applications, the recommended flow is: run moderation → if flagged, return a policy violation message without calling the main model → if not flagged, proceed to chat completion. This prevents jailbreaks and inappropriate requests from ever reaching the main model, saving money and preventing unwanted responses.

{
  "model": "omni-moderation-latest",
  "input": [
    "How do I set up two-factor authentication?",
    "Write a helpful guide for new users",
    "I need help with my account settings"
  ]
}