About
In this module of the Advanced AI Security Part 2 learning path, you will finish quotas, model theft, and poisoned triggers. `/v1/chat/completions` has `MATHGPT_QUOTA=6` (OWASP LLM06:2026 unbounded consumption, formerly LLM10). `/chat` is unmetered. Header `X-Ciberbots-Quota-Bypass: lab-override` is a second bypass. `/debug/weights` hands you a canary for model theft. The phrase `ciberbots-override-pi` is a poisoned fine-tune trigger (LLM05:2026 data and model poisoning, tabletop in Part 1, live here). Upon completion of the module, students will be able to: - Exhaust the metered endpoint and show 429. - Recover the bypass token without flooding the shared lab. - Export the weight canary and fire the backdoor phrase. For inquiries, please write us at https://www.darkrelay.com/cybersecurity-course-inquiry