CY0-001 study resources

SecAI+ is new, so third-party material is thin and some of it is poor. The compensating advantage is that the underlying subject is well documented in free, high-quality public frameworks. CompTIA’s objectives plus the published AI security literature covers this exam better than most paid courses currently do.

Start with the official material

The exam objectives. CompTIA publishes a detailed objectives document. Since the exam is new, this matters more than usual — it is the only authoritative scope, and course authors are still catching up.

Note the weighting. Securing AI systems is 40%, AI-assisted security 24%, AI governance risk and compliance 19%, and basic AI concepts 17%. This is predominantly an offence-and-defence exam, not a governance one.

Free material genuinely worth reading

The AI security field publishes openly, and the primary sources are better than most summaries of them:

• The OWASP Top 10 for LLM applications. Prompt injection, insecure output handling, training data poisoning, model denial of service, excessive agency. This maps closely onto the exam’s largest domain and is free.
• The NIST AI Risk Management Framework. Covers the governance domain properly, and gives you the vocabulary the exam’s GRC questions use.
• MITRE ATLAS. A structured catalogue of adversarial techniques against AI systems — the model extraction, membership inference and poisoning content is directly relevant.

Reading these three properly covers more of the exam than any single course currently available.

Practical work that helps

You do not need a lab in the Network+ sense, but hands-on exposure sharpens the reasoning:

• Try prompt injection against your own test application. Watch retrieved content override an instruction, then fix it by isolating retrieved content from the instruction channel. That is the exam’s favourite mitigation and it becomes obvious once you have done it.
• Scope an agent’s tools deliberately narrowly, then widen them and think about blast radius. Excessive agency is a recurring theme.
• Look at what your logs would actually contain after an incident, and whether that creates a new data risk of its own.

Worth paying for

A practice test with explanations, for diagnosis. The free 20-question test here is weighted to the published domains.

Little else, for now. Given how new the exam is, verify that any paid course was written against the current objectives rather than adapted from a general AI security course.

What to skip

Braindumps. CompTIA revokes certifications over these, and for a new exam the circulating material is likely to be wrong as well as prohibited.

General AI courses. Knowing how models work is 17% of this exam. The other 83% is security reasoning applied to them.

Material that treats this as an entry point into security. It is not. It assumes you can already reason about least privilege, logging and blast radius. If you cannot, start with Network+ and a core security certification.

A sensible order

1. Read the objectives and note that securing AI systems is 40%.
2. Read OWASP’s LLM Top 10, then MITRE ATLAS, then the NIST AI RMF.
3. Try the practical exercises above against your own test application.
4. Take the practice test.
5. Return to the primary sources for the weakest domain.

Manager or practitioner?

If your output is policy and risk decisions rather than configurations, AAISM may fit better. AAISM vs SecAI+ sets out the difference.