TAISE
Free CSA Trusted AI Safety Expert (TAISE) Practice Test
12 exam-style questions with full explanations — no sign-up. Score yourself, then close your gaps with the full course.
Exam fee ~$5956 exam domainsLevel Intermediate2 timed practice tests in the course
✅ Free practice — no sign-up📝 Real exam-style questions💡 Detailed explanations💸 30-day money-back via Udemy
A financial services firm wants to ground an LLM-based assistant in frequently updated internal policy documents without retraining the model. Which approach BEST meets this requirement?
RAG retrieves the most current document content at query time via semantic search, keeping the assistant grounded in up-to-date policy without the cost, latency, or catastrophic-forgetting risk of retraining.
A hiring algorithm is found to produce disparate approval rates against a protected demographic group, even though the protected attribute itself was never used as a model input. Which action BEST addresses this issue while maintaining regulatory compliance?
Disparate impact without an explicit protected attribute indicates proxy discrimination through correlated features. TAISE guidance requires formal bias mitigation (re-weighting, re-sampling, or adversarial debiasing) combined with documented disparate impact testing before redeployment to both fix the problem and demonstrate due diligence.
A financial services company deploys a fraud-detection model as a public-facing API. Attackers systematically query the API with crafted inputs, record outputs, and use the input-output pairs to train a near-identical local copy of the model, avoiding the cost of building it from scratch. Which attack category in the TAISE threat taxonomy does this describe?
Model extraction (also called model stealing) occurs when an adversary uses repeated query-response pairs from an exposed inference API to reconstruct a functionally equivalent copy of the proprietary model, stealing IP without accessing training data or weights directly.
Within the NIST AI Risk Management Framework (AI RMF 1.0), which of the four core functions is responsible for establishing the context of an AI system's risks — including its intended use, impacted stakeholders, and applicable legal/regulatory requirements — before measurement activities begin?
Map is the function that frames the context of a specific AI system — its use case, impacted actors, and risk sources — categorizing risks relative to that context before they are quantified in Measure or treated in Manage.
A company builds a RAG-based customer support assistant that retrieves content from a document store populated by external vendors. An attacker embeds hidden instructions inside a vendor PDF telling the model to exfiltrate conversation history to an external URL. The model complies when that document is retrieved. Which mitigation BEST addresses this attack?
This is indirect prompt injection via retrieved content. The correct control treats all retrieved/external data as untrusted, applies content sanitization/instruction-stripping before it reaches the model context, and restricts what the model is allowed to do (e.g., outbound calls) regardless of instructions embedded in the data.
A company is building an LLM-based chatbot trained on historical customer support tickets, which contain names, emails, and account numbers. The security team must reduce the risk of PII exposure in both the training corpus and model outputs while preserving data utility for fine-tuning. Which approach BEST satisfies TAISE's data security and privacy guidance?
TAISE emphasizes data minimization and privacy-by-design: PII should be identified and redacted/tokenized upstream, before it enters training data, reducing the risk surface at the source rather than relying solely on downstream controls.
Which architectural component is responsible for converting text into fixed-length numerical vectors that capture semantic meaning for similarity search in a RAG pipeline?
Embedding models transform text chunks into dense vector representations that encode semantic meaning, enabling similarity-based retrieval against a vector database.
A bank denies an individual's loan application using a machine learning model. The applicant requests to know why their specific application was rejected. Which approach provides the MOST appropriate explanation for this individual decision?
Individual, instance-level decisions require local explainability techniques. SHAP values decompose a single prediction into per-feature contributions, giving the applicant a concrete, personalized reason for the denial — this is the standard TAISE-recommended approach for adverse action explanations.
During the data collection and preparation stage of the AI lifecycle, an insider with access to the training data pipeline injects a small number of mislabeled images that cause the model to misclassify a specific pattern once deployed. Which lifecycle stage is the PRIMARY point of compromise, and which attack type is this?
Data poisoning is introduced at the data collection/preparation stage by tampering with training data (e.g., mislabeling) so the resulting model learns a hidden flaw or backdoor; the compromise occurs before training even begins.
A retail company deploys a machine learning model to automatically screen and rank job applicant resumes before a human recruiter reviews them. Under the EU AI Act, how should this system be classified?
Annex III of the EU AI Act explicitly lists AI systems used to screen or filter job applications and evaluate candidates as high-risk, triggering obligations like risk management, data governance, and human oversight.
An organization hosts inference endpoints for multiple customers on shared GPU infrastructure in the cloud. A security review requires that one tenant's prompts, cached KV-states, and model weights never be accessible to another tenant, even if one tenant's container is compromised. Which control provides the STRONGEST isolation guarantee for this requirement?
Hardware-level isolation (dedicated GPU instances or confidential computing/TEEs) combined with network and IAM segmentation is the strongest defense against cross-tenant data leakage from compromised containers, since it removes shared-kernel and shared-memory attack surfaces that container namespaces alone cannot fully eliminate.
Several hospitals want to collaboratively improve a diagnostic AI model using patient records from each institution, but regulatory and contractual constraints prohibit any hospital from transmitting raw patient data to a central server. Which privacy-enhancing technique from the TAISE curriculum directly addresses this requirement?
Federated learning trains a shared model by sending model updates (gradients/weights) from each local site to an aggregator, while raw data never leaves the originating institution — directly solving the 'no raw data transfer' constraint.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →TAISE exam — quick answers
How much does the TAISE exam cost?
The exam fee is approximately $595 and varies by region — confirm current pricing with the certification vendor before you book.
What topics are on the exam?
It covers 6 domains: AI and Generative AI Foundations (~18%), AI Ethics, Transparency, and Explainability (~14%), AI Model Lifecycle and Threat Taxonomy (~16%), AI Governance, Risk Management, and Compliance (~16%), AI Security and Safety; Cloud and AI Security (~20%), Data Security and Privacy; Continuous Learning and Adaptation (~16%). The full course has a dedicated chapter, lab and practice-test coverage for each.
Is this practice test really free?
Yes — all questions on this page are free with explanations and no sign-up. The paid Udemy course adds two full-length timed exams, video lessons and hands-on labs.
Will this prepare me for the real exam?
The questions mirror the real exam's style and are mapped to the official domains. This is exam-focused preparation — combine the free test with the full course's timed simulations to gauge your readiness.