Abstract: An API response field classification service obtains API documentation published by a vendor and defined security policies and matches the response fields represented in the security policies to their descriptions in the API documentation. The service generates labelled training data that comprise the identified response field descriptions with labels indicating that their corresponding response field is security related. Additional labelled training data for security unrelated response fields comprises descriptions of response fields that are known not to be represented with any security policies. The service trains a text classifier on the labelled training data. The trained text classifier accepts inputs comprising descriptions of unknown response fields and outputs predicted classes indicating whether the corresponding response fields are predicted to be security related. Subsequent creation of security policies can be focused on these response fields predicted to be security related.