Toxicity in Mask API

How toxicity analysis appears in Identify and Mask (Auto-detect) responses alongside masked entity data.

{
  "mask": [
    {
      "value": "Sarah Connors can be reached at 217-555-2020 or sarah.connors88@example.com"
    }
  ]
}

curl -X PUT https://protecto-trial.protecto.ai/api/vault/mask \
  -H "Authorization: Bearer <AUTH_TOKEN>" \
  -H "Content-Type: application/json; charset=utf-8" \
  -d '{
    "mask": [
      {
        "value": "Sarah Connors can be reached at 217-555-2020 or sarah.connors88@example.com"
      }
    ]
  }'

{
  "data": [
    {
      "value": "Sarah Connors can be reached at 217-555-2020 or sarah.connors88@example.com",
      "token_value": "<PER>6qdEj</PER> can be reached at <PHONE_NUMBER>397-481-2948</PHONE_NUMBER> or <EMAIL_ADDRESS>bqpzj.wfzewdj44@hgbraqmp.pnk</EMAIL_ADDRESS>",
      "individual_tokens": [
        {
          "value": "Sarah Connors",
          "token_value": "6qdEj",
          "entity_type": "PER",
          "start": 0,
          "end": 13
        },
        {
          "value": "217-555-2020",
          "token_value": "397-481-2948",
          "entity_type": "PHONE_NUMBER",
          "start": 32,
          "end": 44
        },
        {
          "value": "sarah.connors88@example.com",
          "token_value": "bqpzj.wfzewdj44@hgbraqmp.pnk",
          "entity_type": "EMAIL_ADDRESS",
          "start": 48,
          "end": 75
        }
      ],
      "toxicity_analysis": {
        "toxicity": 0.0006348,
        "severe_toxicity": 0.0001197,
        "obscene": 0.0001864,
        "threat": 0.0001172,
        "insult": 0.0001784,
        "identity_attack": 0.0001435
      }
    }
  ],
  "success": true,
  "error": { "message": "" }
}

Toxicity analysis is returned when using Identify and Mask (Auto-detect). In addition to masked output, the response includes a toxicity_analysis object and individual_tokens with detected entity positions.

How it appears in the response

Important notes

Toxicity detection does not depend on policy configuration in the Mask API
Masking proceeds normally regardless of toxicity score
Scores reflect the full sentence context, not individual entities
individual_tokens provides entity-level detection results with character positions
Toxicity analysis is not available for token-based or format-based masking — only auto-detect

Toxicity detection always runs on the original text, before any masking occurs. This ensures that masking PII does not affect toxicity scoring.

Was this page helpful?

Last updated 2 days ago

Built with Documentation.AI

{ "data": [ { "value": "Sarah Connors can be reached at 217-555-2020 or sarah.connors88@example.com", "token_value": "<PER>6qdEj</PER> can be reached at <PHONE_NUMBER>397-481-2948</PHONE_NUMBER> or <EMAIL_ADDRESS>bqpzj.wfzewdj44@hgbraqmp.pnk</EMAIL_ADDRESS>", "individual_tokens": [ { "value": "Sarah Connors", "token_value": "6qdEj", "entity_type": "PER", "start": 0, "end": 13 }, { "value": "217-555-2020", "token_value": "397-481-2948", "entity_type": "PHONE_NUMBER", "start": 32, "end": 44 }, { "value": "sarah.connors88@example.com", "token_value": "bqpzj.wfzewdj44@hgbraqmp.pnk", "entity_type": "EMAIL_ADDRESS", "start": 48, "end": 75 } ], "toxicity_analysis": { "toxicity": 0.0006348, "severe_toxicity": 0.0001197, "obscene": 0.0001864, "threat": 0.0001172, "insult": 0.0001784, "identity_attack": 0.0001435 } } ], "success": true, "error": { "message": "" } }