Skip to main content

Run Red-Teaming Scan

AgentSuite-Red runs red-teaming scans end-to-end through a fully UI-driven console — no scripts, no JSONL task files, no CLI required. The same UI is used whether you're driving the SaaS service or a self-hosted on-prem deployment.

The SaaS service is available at https://agentsuite-red.virtueai.cc/. For self-hosted deployments, see On-Prem Deployment — the dashboard behaves identically once it's up.

Login​

Log in using the username and password provided for your corresponding tenant.

Connect your agent​

Go to the Agents tab: https://agentsuite-red.virtueai.cc/agents.

Click "Add Agent" to register your agent, ensuring it complies with the request/response format. Then click "Connect Agent" to test the connection and validate platform compatibility.

Agents tab — Add Agent dialog

Create a red-teaming scan​

Navigate to the New Scan page and follow the step-by-step configuration wizard. The wizard has three stages.

New Scan — start of the wizard

1. Red-teaming environment configuration​

Define the domains and environments in ForgingGround that will interact with your agent during red-teaming. Custom environments can also be integrated via API or MCP.

Environment configuration step

2. Red-teaming objective configuration​

After selecting environments, define the red-teaming objectives by choosing relevant policies and risk categories. You can also upload PDFs containing policies, constraints, and risk definitions, which are automatically extracted to enable customized red-teaming.

Objective configuration step

3. Red-teaming approach configuration​

Configure the threat model and specify allowed attack vectors, such as direct prompt injection or indirect attacks (tool injection, environment injection, skill injection, etc.).

Approach configuration step

Start the red-teaming scan​

Once the scan is created, go to the scan page and click Start to begin. A typical scan runs for 1–2 hours and executes thousands of test instances. You can monitor progress in real time on the scan details page.

Scan list with Start button

Scan running — live progress view

View results​

After completion, analyze results including attack success rates by risk category and review detailed red-teaming trajectories generated during the scan.

Results view with risk-category breakdown

Download report​

Generate and download a comprehensive red-teaming report. The report includes statistical analysis (attack success rates by domain, risk category, and environment) as well as detailed agent failure-mode examples for each category.

Downloadable red-teaming report

Running scans on-prem​

If you need evaluation traffic to stay inside your own network, On-Prem Deployment walks through the full Docker Compose stack — the same UI flow described above runs against your self-hosted backend, and the bundled agentsuite-red CLI is documented there for scripted runs.