AI Red TeamingTargets

Targets Overview

A target is the system you want to test—either a foundational model you connect to directly, or your own agent endpoint.

Choose the right target type

Model

Use Model to quickly test a foundation model directly, with an optional system prompt. No endpoint is required on your side.

Agent

Use Agent to test your own agent endpoint.

Create a target

Creating a target has three parts: connect to the system, optionally share what you already know about it, and review the profile Red builds.

1

Connect your target

Go to Create target (or Targets → New Target in the sidebar). Enter a Target name, choose Model or Agent, and complete the connection. For the exact fields, follow the dedicated guides: Model or Agent. For Agent targets, Red verifies the connection when you continue.

2

Share what you know (optional)

Optionally provide the target’s ground-truth system prompt, tool definitions, and custom refusal phrases. Red verifies what you declare and discovers the rest—the more you share, the sharper the profiling. Click Continue to create the target and start profiling.

3

Review the profile

Red explores the target and builds a profile. Review and edit it, then continue. See Review the profile below.

Review the profile

After the target is created, Red probes it and populates a profile you review once. This profile is stored on the target and reused for every scan against it.

The profile has two parts:

  1. Target Summary - what the system is and who it serves, plus any ground-truth system prompt or tool definitions you provided.
  2. Behavioral rules - the target’s allowed and forbidden actions.

Fields populate as each probe finishes; you can edit any of them. When the profile is ready, click Go to target to open the target, or Run your first scan to go straight to a scan. Both remain disabled until profiling finishes so the profile isn’t saved half-complete.

Profiling is only available for Model and Agent targets. SDK targets are profiled from the SDK, not the platform UI.

The target page

An existing target opens to three tabs:

  1. Scans - scans run against this target, and the entry point to launch a new one.
  2. Profile - the summary and behavioral rules Red built. Edit any field inline; if a target has no profile yet, use Build profile to run profiling.
  3. Connection - the endpoint, authentication, and conversation-handling configuration. Use Edit to change how Red connects.