Access Checks

    Method: aisafety.classifyContent

    • Classify content against a set of predefined safety policies using the classifyContent API.

    • Input content for classification can be provided as text, along with optional context and language.

    • Customize classification by specifying policy types and score thresholds.

    • The API response includes a score and violation result for each policy evaluated.

    • The service uses the latest or a specified version of the classifier for content analysis.

    Analyze a piece of content with the provided set of policies.

    HTTP request

    POST https://checks.googleapis.com/v1alpha/aisafety:classifyContent

    The URL uses gRPC Transcoding syntax.

    Request body

    The request body contains data with the following structure:

    JSON representation 
     { 
 "input" 
 : 
 { 
 object (  InputContent 
 
) 
 } 
 , 
 "context" 
 : 
 { 
 object (  Context 
 
) 
 } 
 , 
 "policies" 
 : 
 [ 
 { 
 object (  PolicyConfig 
 
) 
 } 
 ] 
 , 
 "classifierVersion" 
 : 
 enum (  ClassifierVersion 
 
) 
 }
    Fields
    input

    object ( InputContent )

    Required. Content to be classified.
    context

    object ( Context )

    Optional. Context about the input that will be used to help on the classification.
    policies[]

    object ( PolicyConfig )

    Required. List of policies to classify against.
    classifier Version

    enum ( ClassifierVersion )

    Optional. Version of the classifier to use. If not specified, the latest version will be used.

    Response body

    Response proto for aisafety.classifyContent RPC.

    If successful, the response body contains data with the following structure:

    JSON representation 
     { 
 "policyResults" 
 : 
 [ 
 { 
 object (  PolicyResult 
 
) 
 } 
 ] 
 }
    Fields
    policy Results[]

    object ( PolicyResult )

    Results of the classification for each policy.

    Authorization scopes

    Requires the following OAuth scope:

    • https://www.googleapis.com/auth/checks

    For more information, see the OAuth 2.0 Overview .

    InputContent

    Content to be classified.

    JSON representation 
     { 
 // Union field input 
can be only one of the following: 
 "textInput" 
 : 
 { 
 object (  TextInput 
 
) 
 } 
 // End of list of possible types for union field input 
. 
 }
    Fields
    Union field input . Content to be classified. input can be only one of the following:
    text Input

    object ( TextInput )

    Content in text format.

    TextInput

    Text input to be classified.

    JSON representation 
     { 
 "languageCode" 
 : 
 string 
 , 
 // Union field source 
can be only one of the following: 
 "content" 
 : 
 string 
 // End of list of possible types for union field source 
. 
 }
    Fields
    language Code

    string

    Optional. Language of the text in ISO 639-1 format. If the language is invalid or not specified, the system will try to detect it.

    Union field source . Source of the text to be classified. source can be only one of the following:
    content

    string

    Actual piece of text to be classified.

    Context

    Context about the input that will be used to help on the classification.

    JSON representation 
     { 
 "prompt" 
 : 
 string 
 }
    Fields
    prompt

    string

    Optional. Prompt that generated the model response.

    PolicyConfig

    List of policies to classify against.

    JSON representation 
     { 
 "policyType" 
 : 
 enum (  PolicyType 
 
) 
 , 
 "threshold" 
 : 
 number 
 }
    Fields
    policy Type

    enum ( PolicyType )

    Required. Type of the policy.
    threshold

    number

    Optional. Score threshold to use when deciding if the content is violative or non-violative. If not specified, the default 0.5 threshold for the policy will be used.

    PolicyType

    The unique identifier for a safety policy.

    Enums
    POLICY_TYPE_UNSPECIFIED Default.
    DANGEROUS_CONTENT The model facilitates, promotes or enables access to harmful goods, services, and activities.
    PII_SOLICITING_RECITING The model reveals an individual’s personal information and data.
    HARASSMENT The model generates content that is malicious, intimidating, bullying, or abusive towards another individual.
    SEXUALLY_EXPLICIT The model generates content that is sexually explicit in nature.
    HATE_SPEECH The model promotes violence, hatred, discrimination on the basis of race, religion, etc.
    MEDICAL_INFO The model facilitates harm by providing health advice or guidance.
    VIOLENCE_AND_GORE The model generates content that contains gratuitous, realistic descriptions of violence or gore.
    OBSCENITY_AND_PROFANITY

    ClassifierVersion

    Version of the classifier to use.

    Enums
    CLASSIFIER_VERSION_UNSPECIFIED Unspecified version.
    STABLE Stable version.
    LATEST Latest version.

    PolicyResult

    Result for one policy against the corresponding input.

    JSON representation 
     { 
 "policyType" 
 : 
 enum (  PolicyType 
 
) 
 , 
 "score" 
 : 
 number 
 , 
 "violationResult" 
 : 
 enum (  ViolationResult 
 
) 
 }
    Fields
    policy Type

    enum ( PolicyType )

    Type of the policy.
    score

    number

    Final score for the results of this policy.
    violation Result

    enum ( ViolationResult )

    Result of the classification for the policy.

    ViolationResult

    Result of the classification for the policy.

    Enums
    VIOLATION_RESULT_UNSPECIFIED Unspecified result.
    VIOLATIVE The final score is greater or equal the input score threshold.
    NON_VIOLATIVE The final score is smaller than the input score threshold.
    CLASSIFICATION_ERROR There was an error and the violation result could not be determined.
    Create a Mobile Website
    View Site in Mobile | Classic
    Share by: