To utilize the AI Red Teaming service, you must create and register your AI targets in the Netskope UI.

To create an AI target:

1. Go to AI Red Teaming.

2. Click the Target Registry tab.

3. Click New Target and choose one of the following methods to connect to a target:

   Rest API

   LLM

   OpenAI Compatible API

   Integrate any service for testing as long as it offers a REST API, establishes a successful connection, and meets all remaining requirements. This flexibility allows you to connect to AI models or agents they have built themselves or to other AI vendors that AI Red Teaming doesn’t officially support yet (e.g., Google Gemini API).

   • Name: Enter a name for your target.

   • Description: (Optional) Enter comments or notes for the target.

   • Connect Type: REST API is automatically selected.

   • URL: Enter the URL of the target. For example:

     https://api.openai.com/v1/responses

   • Request Template: Enter the response body to the target. It must includes the keyword {prompt}. This is for injecting red teaming prompt when Netskope attacks the target.

     For example:

     {
         "model": "gpt-5-nano",
         "input": "{prompt}"
     }

   • Response Path: Enter the JSON path for parsing the target response from the REST API. For example, if the target REST API responds with {"response": "I cannot answer your question"}, set response as the response path. This way Netskope can extract the LLM response “I cannot answer your question” from the REST API response.

   • HTTP Headers: Choose the key and enter the value of the header. You can enable Obfuscation if the value of the header is sensitive. If you want to add multiple HTTP headers, click +.

   • Request per Minute: Enter the number of requests to send to the target within a one-minute interval. The default is 60.

     

     Streaming Responses (SSE)

     AI Red Teaming automatically handles streaming responses using Server-Sent Events (SSE). When the target API returns a Content-Type: text/event-stream response, Netskope detects this and assembles the final response from the stream. No additional configuration is required.

     Enabling SSE on an OpenAI-Compatible Endpoint

     For LLM servers that support both streaming and non-streaming modes (such as OpenAI, LM Studio, and Ollama):

     1. Enter "stream": true in Request Template. Example Request Template:

        {
          "model": "{model}",
          "messages": [{"role": "user", "content": "{prompt}"}],
          "stream": true
        }

     2. Set the Response Path to match the streaming response format. Example Response Path:

        choices[0].delta.content

     Servers That Always Stream

     If the target server always returns SSE responses, then you don’t need to change the Request Template. Set the Response Path to the JSON field that contains the response text in each chunk.

     Supported SSE Formats

     Format	Example Response Path
     OpenAI Chat Completions (stream: true)	choices[0].delta.content
     OpenAI Responses API	delta
     Anthropic Messages API	delta.text
     Custom/Internal LLM APIs	Depends on the server’s payload structure

   Connect seamlessly with a specialized LLM provider of a major cloud or AI vendor, such as OpenAI, Amazon Bedrock, and Azure OpenAI.

   • Name: Enter a name for your target.

   • Description: (Optional) Enter comments or notes for the target.

   • Connect Type: LLM is automatically selected.

   • LLM: Choose from one of the following large language models.

     • OpenAI

     • Azure OpenAI

     • Amazon Bedrock

   • API Key: Enter the API Key for the LLM

   • Model: Enter the model name for the LLM.

   • Request per Minute: Enter the number of requests to send to the target within a one-minute interval. The default is 60.

     

   Integrate using the OpenAI Compatible API standard so you can easily connect to any vendor or model that follows the OpenAI specification without requiring a custom configuration.

   • Name: Enter a name for your target.

   • Description: (Optional) Enter comments or notes for the target.

   • Connect Type: OpenAI Compatible API is automatically selected.

   • URL: Enter the base URL of the OpenAI Compatible API. For example:

     https://ai-mock-server.execute-api.us-west-2.amazonaws.com/v1

   • Model: Enter the model name that will be sent to the OpenAI Compatible API endpoint.

   • HTTP Headers: Choose the key and enter the value of the header. You can enable Obfuscation if the value of the header is sensitive. If you want to add multiple HTTP headers, click +.

   • Request per Minute: Enter the number of requests to send to the target within a one-minute interval. The default is 60.

     

4. Click Create & Verify to check the target connectivity. If the verification fails, you can still save the target, but it’ll be tagged as Unverified.

Target Registry

In the Target Registry tab, you can see the following information for each registered target:

• Name: The name of your target. Click to modify its settings.

• Tests: View the number of tests performed for the target.

• Verified: View if the target is successfully verified or not.

You can also click the icon and perform any of the following actions:

• Edit: Modify the target and its settings.

• New Test: Takes you to Test Rounds so you can perform a new test on the target.

• Delete: Delete the target.