ROT HACKERSACADEMY
Menu
Start learning
Courses/Basilisk AI Red Teaming/Lesson 02
10.02 / 40 min

Run the ground-truth lab

Start the Docker-isolated Basilisk lab on loopback and verify its secure and vulnerable OpenAI-compatible routes.

By the endInspect the safety boundaryStart the lab on loopbackVerify paired endpoints
Controlled exercise

Use only the course lab on 127.0.0.1:8765. Do not copy these tests to a public system without explicit written authorization.

01

Download and inspect

Download the Academy package, extract it, and read README.md plus docker-compose.yml. Confirm the port publishes only on 127.0.0.1 and the container drops Linux capabilities.

/downloads/basilisk-ground-truth-lab.zip
Expected result

The Compose mapping is 127.0.0.1:8765:8765 and the service uses a read-only filesystem with resource limits.

02

Start the lab

Run Docker Compose from the extracted directory. Keep the terminal visible and do not add a public bind address or tunnel.

docker compose up --build
Expected result

The service listens at http://127.0.0.1:8765 only through the loopback publish rule.

03

Check vulnerable and secure routes

Open the lab root or send harmless requests to both OpenAI-compatible endpoints. The deterministic responses provide a controlled positive and negative control.

POST http://127.0.0.1:8765/vulnerable/v1/chat/completions
POST http://127.0.0.1:8765/secure/v1/chat/completions
Expected result

Both routes respond locally; their ground-truth labels differ by design.

04

Know what is simulated

The SSRF, SQL, and shell scenarios are inert simulations. They do not perform real network calls, database queries, filesystem access, or process execution.

MASTERY CHECKPOINT

Prove it before moving on.

You can prove the lab is loopback-only, name both control routes, and explain why its dangerous-looking behaviors are simulations.