Red Team an Assistant

1. Navigate to the assistants section from the sidebar#

2. Click on the three dots next to your assistant's name.#

3. Click on Red Team#

4. Click on Get Started#

5. Select Target Dataset#

6. If you want to create a custom dataset: Go to the Custom Datasets option.#

7. Drag and drop your dataset file here.#

8. Here, you will see all of your custom datasets listed.#

9. If you want to use a built-in dataset, Click on the Built-in Datasets option.#

10. Select the appropriate category for your dataset.#

11. Click on Next#

12. Provide a clear and descriptive name for the job, specify the number of prompts to execute#

13. Select Attack Type#

14. Define the attack parameters by selecting the specific methods you want to apply and Click on Next.#

15. Select the Converter that defines how prompts and responses will be transformed or formatted before being sent to the model.#

16. Click on Next#

17. Here you can review your configuration before launching job.#

18. Click on Launch Job#

19. A new job will appear in the jobs table with status Pending. After a short time, the status will update to Running as the attack is executed.#

20. Click on View Attack Details to review your configuration to ensure they match your testing goals.#

21. A modal will open where you can view the complete details of your attack.#

22. If your attack is in Running state and you want to stop it, click on Cancel Attack.#

23. Click on Cancel Attack. Remember that this action can't be undone.#

24. Once the attack completes, the job status will change to Completed. Click on View Result#

25. Here you can inspect the prompts, model responses, and Scorer outputs for each red team example.#

26. Click on Export Vulnerability Report to export a report with vulnerable prompts along with model response and rationale.#