CAICSO

The exam is multiple-choice, delivered in English, and covers the eight published domains at their stated weights. The official body publishes the current question count, time limit, and delivery options (test center or remote proctored), so check its site for the authoritative numbers rather than relying on forum posts.

Most candidates with a security or GRC background find the governance and risk domains approachable and the adversarial ML and MLOps domains harder, because they require knowing how attacks work technically, not just their names. Scenario questions reward people who have seen real AI deployments. Expect a moderate-to-hard exam that punishes pure memorization.

A common plan is eight to twelve weeks at five to eight hours per week for someone new to AI security, shorter for practitioners. Spend roughly half the time on the three 15% technical domains and the rest on governance and the smaller domains. Build a one-page control map per framework as you go.

Start with the threat landscape and risk identification so later controls have context, then security architecture, data security, and lifecycle security. Study governance frameworks after you understand the risks they address, and finish with incident response and program leadership. This order makes framework clauses feel like answers to problems you already know.

No. You need to understand model artifacts, training and inference pipelines, data flows, and where trust boundaries sit, but not how to tune a model. Being able to read a pipeline diagram, identify where a model is fetched from, and explain what telemetry exists is far more useful than coding ability.

Candidates lose points by confusing classic application security with AI-specific risk, treating prompt injection as a solved input-validation problem, and memorizing framework names without knowing what each control actually mitigates. Another frequent error is ignoring supply chain risk in models and datasets. Read every scenario for who the attacker is and what they gain.

Retake rules, waiting periods, and fees are set by the certifying body and can change, so confirm the current policy on its official site before scheduling. In practice, use a score report to identify weak domains, revisit those with hands-on practice, and avoid rebooking immediately without changing your study approach.

Recertification typically involves continuing education credits or a renewal fee, and the exact requirement and cycle length are published by the certifying body. Check its official site for the current validity period and credit rules, and track your activities as you earn them so renewal is not a scramble.

Use the primary framework documents themselves: the NIST AI Risk Management Framework, ISO/IEC 42001 and 23894, and the OWASP LLM and ML security top-ten lists. Add vendor-neutral threat modeling practice on a RAG or classification system. Avoid relying on brain dumps, which are often outdated and teach the wrong reasoning.