Specialty · Specialty exam · ASC
AI Security Specialist
Validates that you can identify the threats specific to LLM applications and agents (prompt injection, data leakage, excessive permissions), defend against and detect them, and verify your defenses.
Daily practice · Free
10 questions · 25 min · No sign-up needed
Specialty exam · Proctored
35 questions · 60 min · Raw score of 75 or more · $3.99 / one attempt
Specialty exam detailsExam overview
- Questions
- 35
- Time
- 60 min
- Pass mark
- 75 pts
- Recommended experience
- Red Belt (LLM API Developer) level and basic knowledge of application security
Every question has five choices and is marked on an OMR answer sheet. Multiple-choice and true-or-false questions ask for the right combination from a box; ordering and matching questions ask for the right arrangement. Judgment questions (L3) are worth 3 points and all others 2; the raw score is converted to a 100-point scale and graded 1 to 9.
Exam domains
1. Threat modeling20%
2. Prompt injection25%
3. Data protection and permissions20%
4. Defense testing and red teaming20%
5. Operational security and incident response15%
Weights follow the Specialty exam. Practice and mock exams may mix domains differently.
What this exam verifies
- Classify threats to LLM applications using the OWASP LLM Top 10
- Find and mitigate paths for direct and indirect prompt injection
- Prevent leakage by minimizing tool permissions and data flows
- Verify defenses with guardrails, monitoring and red teaming
- Design incident response and log retention
Who it is for
- Security reviewers for AI products
- Security engineers
- Agent platform developers
Does not go deep into general network security or research on attacks against model training.
Domain details
Domain 1. Threat modeling
- Knowledge: OWASP LLM Top 10, Trust boundaries, Assets and attack surface
- Skills: Find trust boundaries in a data flow diagram
Domain 2. Prompt injection
- Knowledge: Direct and indirect injection, Injection via documents, web pages and tool results, How it differs from jailbreaking
- Skills: Identify injection paths, Combine mitigations (privilege separation, confirmation, output restrictions)
Domain 3. Data protection and permissions
- Knowledge: Sensitive information disclosure, Excessive agency, Improper output handling
- Skills: Minimize tool permissions, Render output safely
Domain 4. Defense testing and red teaming
- Knowledge: Guardrails, Classifiers, Red-teaming methods, Evaluation sets
- Skills: Design attack scenarios, Measure how well defenses work
Domain 5. Operational security and incident response
- Knowledge: Audit logs, Cost and resource exhaustion attacks, Supply chain (models and plugins)
- Skills: Establish incident response procedures
Official documentation
OpenAI
Specialty Course · AI Security Specialist
Practice, mock exams and every course book are free. Pay only when you take an official test.