About Infirmary

Free, original AI red-teaming education by Xtrinel

Our Mission

Infirmary exists to make AI security education accessible to everyone. We believe that understanding how to attack AI systems is fundamental to defending them effectively.

Built by Xtrinel, a professional offensive security company, Infirmary distills years of real-world AI red-teaming experience into 29 hands-on labs covering prompt injection, data poisoning, evasion attacks, privacy attacks, and agent exploitation.

What You'll Learn

Prompt Injection Attacks

Direct injection, delimiter tricks, encoding bypasses, RAG poisoning, and escalation to RCE

Data Attacks

Label flipping, clean-label backdoors, supply-chain model poisoning

Evasion Techniques

Gradient-based adversarial examples (FGSM, DeepFool), sparse perturbations (JSMA)

Agent Exploitation

Tool misuse, privilege escalation, multi-step attack chains, human-in-the-loop bypasses

Privacy Attacks

Membership inference, differential privacy defenses, privacy-accuracy tradeoffs

Defense Evasion

Guardrail bypasses, benchmark gaming, evaluation brittleness

Lab Format

Infirmary offers two types of labs:

Live Target
Interactive web services you'll attack in real-time. These labs provision ephemeral containers running vulnerable AI systems. Your goal: exploit them to extract flags.
Notebook
Jupyter notebooks with guided exercises. Build classifiers, poison datasets, craft adversarial examples, and measure attack effectiveness through code.

VAAST Integration (Optional)

All Infirmary labs can be solved manually. For those seeking automated workflows, VAAST (Vulnerability Assessment for AI Security Testing) provides payload generation and attack automation.

VAAST Free covers basic prompt injection and encoding bypasses. VAAST Pro adds MCP scanning, RAG poisoning, and gradient-based evasion generation. SentrinelNet's eight offensive agents validate multi-step attack chains in advanced labs.

Who Built This

Infirmary is built and maintained by Xtrinel, a professional offensive security company specializing in AI red-teaming.

The curriculum draws from HYDRACUDA, Xtrinel's research division, which has published original research on prompt injection, agent exploitation, and adversarial machine learning.

Ready to Learn AI Red-Teaming?

Get Started