Loading demo links...
What to study for each 312-52 domain
AI Attack Surface and Threat Landscape
Maps the components, data flows, and trust boundaries that expose AI systems to attack. Covers OWASP LLM risks, MITRE ATLAS reconnaissance and discovery, and enterprise threats such as shadow AI.
Key skills
AI threat modeling, OWASP LLM Top 10, MITRE ATLAS, attack surface mapping, RAG and vector database risks, risk prioritization
Prompt Injection and Jailbreaking
Covers direct and indirect prompt injection, prompt leaking, and jailbreak methods from role-play and encoding to many-shot and multi-turn escalation. Also covers layered defenses including delimiting, output validation, and privilege separation.
Key skills
Direct injection, indirect injection, prompt leaking, jailbreak techniques, token smuggling, Crescendo attacks, injection defenses, automated red teaming
Adversarial Machine Learning and Model Attacks
Examines attacks against models and their training pipelines, including extraction, inversion, membership inference, data poisoning, backdoors, and evasion. Also covers defenses such as adversarial training, query monitoring, and output limiting.
Key skills
Model extraction, model inversion, membership inference, data poisoning, backdoor attacks, FGSM evasion, adversarial training, watermarking
Agentic AI and Supply Chain Attacks
Focuses on autonomous agents, tool use, memory, and multi-agent trust, plus the AI supply chain of models, datasets, plugins, and libraries. Covers Excessive Agency, unsafe serialization, and AI-BOM practices.
Key skills
Excessive Agency, tool-use hijacking, memory poisoning, model supply chain security, plugin risk, AI-BOM, scoped credentials, integrity verification
AI Security Testing, Hardening and Incident Response
Covers structured AI penetration testing, guardrail and regression testing, and hardening for LLM and RAG systems. Also covers AI-specific incident containment, evidence preservation, rollback, and remediation.
Key skills
AI penetration testing, guardrail testing, RAG hardening, regression testing, AI incident response, canary tokens, model rollback, extraction re-testing