Chapter 2 of 4 · CompTIA SecAI+ CY0-001

Securing AI Systems

Threat modeling, guardrails and gateways, access, data protection, monitoring, attacks and controls.

  • 7topics
  • 39 minfast track
  • 3diagrams and exhibits
  • 40%of the exam
Start with AI Threat-Modeling Resources

Topics in this chapter

Take them in order, or jump to the one you need.

  1. 1AI Threat-Modeling ResourcesUse OWASP Top 10 for LLM and ML Security Top 10, MIT AI Risk Repository, MITRE ATLAS, CVE AI Working Group and STRIDE to threat…Diagram6 min
  2. 2Model Guardrails and AI Gateway ControlsImplement AI security controls: model evaluation, guardrails, prompt templates, prompt firewalls, rate, token, quota and modality…Diagram5 min
  3. 3Access Controls for Models, Data, Agents and APIsApply access controls to AI systems: model access, training and RAG data access, least privilege agent access, and network and…Diagram6 min
  4. 4Encryption and Data Safety Controls for AIProtect AI data with encryption in transit, at rest and in use, plus anonymization, classification labels, redaction, masking and…5 min
  5. 5Monitoring and Auditing AI SystemsMonitor AI prompts, responses, logs, confidence, request rates and cost, and audit models for hallucinations, accuracy, bias…5 min
  6. 6Attacks on Models and Training DataSpot backdoor, trojan, poisoning, inversion, model theft, supply chain, transfer learning, skewing and membership inference…6 min
  7. 7Attacks on LLM Applications and Compensating ControlsAnalyze prompt injection, jailbreaking, insecure output handling, model DoS, plug-in flaws, excessive agency and overreliance…6 min

About this chapter

What this chapter covers

Securing AI Systems is 40% of the CompTIA SecAI+ CY0-001 exam, the largest domain by far. It covers how a security professional protects artificial intelligence (AI) models, the data they use and the applications and agents built on them.

It covers threat modeling (the Open Worldwide Application Security Project (OWASP) Top 10 for large language models (LLM Top 10) and Machine Learning (ML) Security Top 10, the Massachusetts Institute of Technology (MIT) AI Risk Repository, MITRE Adversarial Threat Landscape for Artificial-Intelligence Systems (ATLAS), the Common Vulnerabilities and Exposures (CVE) AI Working Group and general threat-modeling frameworks), security controls (model guardrails, prompt templates, gateway controls such as prompt firewalls, rate, token and modality limits, and guardrail testing), access controls for models, data, agents and application programming interfaces (APIs), data security (encryption in transit, at rest and in use, anonymization, classification, redaction, masking and minimization), monitoring and auditing (prompts, logs, confidence, rate, cost, hallucinations, bias and access) and attacks with compensating controls (from prompt injection and poisoning to model theft, excessive agency and overreliance).

The exam has a maximum of 60 multiple-choice and performance-based questions in 60 minutes. Questions are vendor-neutral and usually close with "Which of the following ...?". In this domain many questions give a scenario, a log excerpt or a set of requirements and ask for the attack name or the best control.

How the topics fit together

Threat modelwhat can go wrongControlsguardrails and gatewayAccesswho and what can actDataencrypt and reduceMonitorprompts, logs, costAttacksevidence and compensating controls

Threat modeling tells you which controls you need, and the control pages build up layer by layer around the model. The two attack pages come last because every compensating control they recommend is one you will already have met.

Exam tip

For any attack question, ask two things: at which point did the attacker get in (training data, model file, prompt, plug-in, output), and which control works at that same point.

How to use these pages

Each topic page has the same shape, so you can skim for what you need.

  • What it is: the idea in plain words.
  • How it works: the mechanism, with tables and code examples.
  • What to remember for the exam: the facts to memorise and how questions are phrased.
  • Common mistakes: the classic traps.
  • Check yourself: three questions to test your understanding.

Ready for more practice?

The full CompTIA SecAI+ CY0-001 question bank has 654 questions, with timed practice exams and an explanation for every answer.

SubscribeTry Sample Exam

This study guide is independent and is not affiliated with, endorsed by, or sponsored by CompTIA. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.