AI Guardrails:Multimodal API integration guide

ServiceParameters

At least one of content, imageUrls, or fileUrls is required per request.

accountId, sessionId, and chatId all identify content across requests, but they are not interchangeable. The following table compares their purposes and scenarios.

Data

Detail

Each item in Detail represents the result for one protection dimension.

Result

Ext

CustomizedHit

Request example

{
  "Service": "XXX",
  "ServiceParameters": {
    "content": "This thermos cup supports IP67 waterproofing and can be used underwater.",
    "chatId": "ABC123",
    "dataId": "img123******",
    "accountId": "abc",
    "sessionId": "abc",
    "imageUrls": ["http://xxxx"], # Only one image is currently supported
    "fileUrls": ["http://xxxx"], # Only one file is currently supported
    "referer": "http://www.aliyun.com",
    "referenceContent": "Product name: Smart thermos cup. Material: 316 stainless steel. Heat retention: 12 hours. Capacity: 500 ml. Package contents: 1 thermos cup, 1 user manual."
  }
}

Response examples

`query_security_check` — system policy matched

The response below shows a request that triggered multiple protection dimensions: a malicious URL, sensitive data masking, a prompt attack, a custom label, a content compliance violation, and a model hallucination.

{
  "Code": 200,
  "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****",
  "Message": "OK",
  "Data": {
    "Suggestion": "block",
    "Detail": [
      {
        "Suggestion": "mask",
        "Type": "sensitiveData",
        "Level": "S2",
        "Result": [
          {
            "Ext": {
              "Desensitization": "...[mobile phone number] is my contact information...",
              "SensitiveData": [
                "136********"
              ]
            },
            "Description": "Mobile phone number (the Chinese mainland)",
            "Label": "1814",
            "Level": "S2"
          },
          {
            "Ext": {
              "SensitiveData": [
                "**City"
              ]
            },
            "Description": "City (the Chinese mainland)",
            "Label": "1739",
            "Level": "S0"
          }
        ]
      },
      {
        "Suggestion": "block",
        "Type": "promptAttack",
        "Level": "high",
        "Result": [
          {
            "Description": "Refusal Suppression Jailbreak",
            "Confidence": 100,
            "Label": "Refusal Supression Jailbreak",
            "Level": "high"
          }
        ]
      },
      {
        "Suggestion": "block",
        "Type": "contentModeration",
        "Level": "high",
        "Result": [
          {
            "Description": "Suspected political entity",
            "Confidence": 100,
            "Label": "political_entity",
            "Level": "high"
          },
          {
            "Ext": {
              "CustomizedHit": [
                {
                  "LibName": "Needs to be blocklisted",
                  "KeyWords": "word_a,word_b,word_c"
                }
              ]
            },
            "Description": "Hit custom library",
            "Confidence": 100,
            "Label": "customized",
            "Level": "high"
          }
        ]
      },
      {
        "Result": [
          {
            "Description": "Intrinsic Hallucination",
            "Confidence": 95,
            "Label": "Intrinsic Hallucination",
            "Level": "medium"
          }
        ],
        "Type": "modelHallucination",
        "Suggestion": "block",
        "Level": "medium"
      }
    ]
  }
}

`img_response_security_check` — digital watermark detected, content compliance violation

{
  "Code": 200,
  "Data": {
    "Detail": [
      {
        "Level": "none",
        "Result": [
          {
            "Confidence": 0.0,
            "Description": "No risk detected",
            "Ext": {
              "FileUrl": "https://sase-public-server-files.oss-cn-hangzhou.aliyuncs.com/saas-XXX",
              "OutFileSize": 527918,
              "FileUrlExp": "1754200240",
              "Filename": "wJGz6kmZ1Ce.jpg",
              "OutFileHashMd5": "02f5129f606027c7a87b84377ec98f8e"
            },
            "Label": "nonLabel",
            "Level": "none"
          }
        ],
        "Suggestion": "pass",
        "Type": "waterMark"
      },
      {
        "Level": "high",
        "Result": [
          {
            "Confidence": 90,
            "Description": "Violation of advertising law - superlative words",
            "Label": "ad_Compliance_WordLimit_Tii",
            "Level": "high"
          }
        ],
        "Suggestion": "block",
        "Type": "contentModeration"
      }
    ],
    "Suggestion": "block"
  },
  "Msg": "OK"
}

`text_img_security_check` — content compliance violation

{
  "Code": 200,
  "Data": {
    "Detail": [
      {
        "Ext": {},
        "Level": "high",
        "Result": [
          {
            "Confidence": 98.34,
            "Description": "Female cleavage",
            "Label": "sexual_Cleavage",
            "Level": "high"
          }
        ],
        "Suggestion": "block",
        "Type": "contentModeration"
      }
    ],
    "Suggestion": "block"
  },
  "Msg": "OK"
}

`file_security_sync_check` — malicious file detected

{
  "Code": 200,
  "Data": {
    "Detail": [
      {
        "Ext": {},
        "Level": "high",
        "Result": [
          {
            "Confidence": 100,
            "Description": "Web shell",
            "Label": "WebShell",
            "Level": "high"
          }
        ],
        "Suggestion": "block",
        "Type": "maliciousFile"
      },
      {
        "Ext": {
          "PageSum": 1
        },
        "Level": "none",
        "Result": [
          {
            "Description": "No risk detected",
            "Label": "nonLabel",
            "Level": "none"
          }
        ],
        "Suggestion": "pass",
        "Type": "contentModeration"
      }
    ],
    "Suggestion": "block"
  },
  "Msg": "OK"
}

`text_file_sec_sync_check` — no risk detected

{
  "Code": 200,
  "Data": {
    "Detail": [
      {
        "Ext": {
          "PageSum": 4
        },
        "Level": "none",
        "Result": [
          {
            "Description": "No risk detected",
            "Label": "nonLabel",
            "Level": "none"
          }
        ],
        "Suggestion": "pass",
        "Type": "contentModeration"
      }
    ],
    "Suggestion": "pass"
  },
  "Msg": "OK"
}

Model hallucination detection examples

Model hallucination detection (public preview) compares large language model (LLM) generated content against the reference content that you pass in referenceContent to identify false or inaccurate information in the model output. The following examples cover three typical hallucination scenarios and one contrast scenario without hallucination. Each example includes the scenario description, the input (submitted content and referenceContent), the detection result, and an explanation. You can copy the sample content to verify your integration. Prerequisites: you enabled the Model Hallucination protection dimension for the service in Configure check items, and you pass referenceContent in ServiceParameters.

Example 1: Information fabricated beyond the reference content

Scenario: the reference content does not mention waterproofing, but the model output fabricates an "IP67 waterproofing" claim.

Input:

{
  "Service": "query_security_check",
  "ServiceParameters": {
    "content": "This thermos cup supports IP67 waterproofing and can be used underwater.",
    "referenceContent": "Product name: Smart thermos cup. Material: 316 stainless steel. Heat retention: 12 hours. Capacity: 500 ml. Package contents: 1 thermos cup, 1 user manual."
  }
}

Detection result:

{
  "Code": 200,
  "Message": "OK",
  "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****",
  "Data": {
    "Suggestion": "block",
    "Detail": [
      {
        "Type": "modelHallucination",
        "Suggestion": "block",
        "Level": "medium",
        "Result": [
          {
            "Description": "Intrinsic Hallucination",
            "Confidence": 95,
            "Label": "Intrinsic Hallucination",
            "Level": "medium"
          }
        ]
      }
    ]
  }
}

Explanation: the "IP67 waterproofing" claim in the submitted content has no basis in the reference content and is judged to be a model hallucination. The response contains a detection item whose Type is modelHallucination in Detail, the Label is Intrinsic Hallucination, the Confidence is the confidence level of the hallucination judgment, and the suggested action is block.

Example 2: Information that contradicts the reference content

Scenario: the reference content states that 7-day no-reason returns are supported, but the model output claims that returns are not supported.

Input:

{
  "Service": "query_security_check",
  "ServiceParameters": {
    "content": "This product does not support returns and cannot be refunded after purchase.",
    "referenceContent": "After-sales service: This product supports 7-day no-reason returns. The buyer pays the return shipping fee."
  }
}

Detection result:

{
  "Code": 200,
  "Message": "OK",
  "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****",
  "Data": {
    "Suggestion": "block",
    "Detail": [
      {
        "Type": "modelHallucination",
        "Suggestion": "block",
        "Level": "medium",
        "Result": [
          {
            "Description": "Intrinsic Hallucination",
            "Confidence": 95,
            "Label": "Intrinsic Hallucination",
            "Level": "medium"
          }
        ]
      }
    ]
  }
}

Explanation: the submitted content directly contradicts the return policy in the reference content and is judged to be a model hallucination (Label: Intrinsic Hallucination). The suggested action is block.

Example 3: Information distorted or exaggerated from the reference content

Scenario: the reference content states a battery life of 8 hours for the standard edition, but the model output exaggerates it to "up to 24 hours of battery life for all models and a full charge in 5 minutes".

Input:

{
  "Service": "query_security_check",
  "ServiceParameters": {
    "content": "All models offer up to 24 hours of battery life and a full charge in just 5 minutes. We recommend the top-tier edition for every user.",
    "referenceContent": "Product specifications: The standard edition has 8 hours of battery life. In fast-charging mode, it charges to 80% in 30 minutes."
  }
}

Detection result:

{
  "Code": 200,
  "Message": "OK",
  "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****",
  "Data": {
    "Suggestion": "block",
    "Detail": [
      {
        "Type": "modelHallucination",
        "Suggestion": "block",
        "Level": "medium",
        "Result": [
          {
            "Description": "Intrinsic Hallucination",
            "Confidence": 95,
            "Label": "Intrinsic Hallucination",
            "Level": "medium"
          }
        ]
      }
    ]
  }
}

Explanation: the submitted content distorts and exaggerates the battery life and charging data in the reference content and is judged to be a model hallucination (Label: Intrinsic Hallucination). The suggested action is block.

No hallucination detected (contrast)

Scenario: the model output is consistent with the reference content, and no hallucination is detected. The response is normal (Code 200), but no detection item whose Type is modelHallucination appears in Detail.

Input:

{
  "Service": "query_security_check",
  "ServiceParameters": {
    "content": "This thermos cup keeps drinks warm for 12 hours and has a capacity of 500 ml.",
    "referenceContent": "Product name: Smart thermos cup. Material: 316 stainless steel. Heat retention: 12 hours. Capacity: 500 ml. Package contents: 1 thermos cup, 1 user manual."
  }
}

Detection result:

{
  "Code": 200,
  "Message": "OK",
  "RequestId": "AAAAAA-BBBB-CCCCC-DDDD-EEEEEEEE****",
  "Data": {
    "Suggestion": "pass",
    "Detail": []
  }
}

Explanation: the submitted content is faithful to the reference content, so it is not judged to be a hallucination and no modelHallucination detection item appears in Detail. The response looks the same when the model hallucination protection dimension is enabled but referenceContent is not passed: the request succeeds, but no hallucination detection result is returned. If you see no modelHallucination item, first check that your request passes referenceContent.