This is not a replacement for the official study guide. Please read the official study guide first — use this exam prep only to reinforce or practice your understanding.
Free AAISM exam practice questions with answers and explanations, organized by domain and part.
Domain 1: AI Governance and Program Management
Q281 A new AI inventory initiative is met with employee reluctance to disclose unapproved tools. Which communication is MOST effective?
- A. Keep the purpose confidential until all of the tools have been fully identified
- B. Announce that departments found using unapproved tools will lose part of their budget
- C. Ask only senior managers to report, since staff rarely use AI tools directly
- D. State that the inventory identifies AI in use and supports governance, with no reprimands
Answer: D
The purpose is to identify AI systems in use, not to penalize anyone, and communicating this encourages disclosure. Threats drive usage underground, and secrecy defeats collaboration. Disclosure supports risk management.
Domain 1: AI Governance and Program Management
Q282 Which factor MOST complicates inventorying an AI system compared with a traditional application such as a spreadsheet tool?
- A. One system may have multiple owners, models, datasets, versions and licenses
- B. AI systems normally rely on one dataset and a single algorithm per release
- C. AI systems are exempt from legal and regulatory requirements until deployed
- D. AI systems rarely use third-party tools, so ownership is almost always single
Answer: A
An AI solution is a composite of many systems with multiple owners, datasets, algorithms and licenses, unlike a single application. AI often uses third-party tools and is subject to compliance rules. The other options reverse reality.
Domain 1: AI Governance and Program Management
Q283 Which artifact should an enterprise that builds its own AI products BEST create and maintain in addition to the general inventory?
- A. A hardware depreciation schedule
- B. A staff holiday roster
- C. An AI model catalog
- D. A vendor payment calendar
Answer: C
Enterprises developing their own AI products or services should create and maintain an AI model catalog. Payment calendars and depreciation schedules are finance records that do not describe models. The catalog supports the inventory and documentation.
Domain 1: AI Governance and Program Management
Q284 What is the LEAST frequent update cycle acceptable for an AI inventory?
- A. Only when an incident occurs
- B. Every ten years
- C. At least annually
- D. Only when regulators request it
Answer: C
The manual states the inventory should be updated at least annually. Waiting for incidents or regulators leaves gaps in visibility. Faster updates may be needed in higher-risk environments.
Domain 1: AI Governance and Program Management
Q285 A security team wants to discover unmanaged AI tools in use. Which method is MOST useful for building a baseline?
- A. Counting managed printers and monitors on the corporate network
- B. Dataflow diagrams, metadata and user access management records
- C. Surveying competitors about which AI tools they currently use
- D. Reviewing vendor brochures and public announcements about AI products
Answer: B
IT and security can use data management and dataflow diagrams, metadata and directory or access management systems for discovery. Brochures and printer counts do not reveal AI use. Competitor surveys are irrelevant to internal discovery.
Domain 1: AI Governance and Program Management
Q286 A data team ranks datasets by the processes they support, using business impact analyses. What is it PRIMARILY identifying?
- A. Critical data elements
- B. Public data
- C. Redundant backups
- D. Obsolete archives
Answer: A
Critical data elements are identified by reviewing BIAs, because data needed by the most critical processes is critical. Public data is a security classification instead. Backups and archives are storage concepts.
Domain 1: AI Governance and Program Management
Q287 A model for public use was trained on confidential customer data. What is the GREATEST risk?
- A. The model will run noticeably slower because confidential data increases its size
- B. The model will require much less testing since the training data is genuine
- C. The training data will become public domain once the model is released
- D. Sensitive data embedded in the model may be disclosed to users without rights to it
Answer: D
Sensitivity of training data must align with the rights of users who access the model, otherwise embedded sensitive data may leak. Speed and testing effort are not the concern. Data does not become public automatically.
Domain 1: AI Governance and Program Management
Q288 Restricted data is copied into a data lake together with data of lower sensitivity, and access controls are not carried over. What is the MOST likely result?
- A. Metadata becomes more accurate because data is gathered in one place
- B. Data once restricted at the source becomes available to more users
- C. The commingled data is automatically encrypted at the lake boundary
- D. Query performance degrades because classification labels multiply
Answer: B
When data from different classifications is commingled without preserved access controls, previously restricted data may be widely available. Encryption does not appear automatically. Metadata may be lost, not improved.
Domain 1: AI Governance and Program Management
Q289 What is the MAIN reason vector databases require new access control models?
- A. They cannot hold metadata, so access must be enforced entirely outside the database
- B. They store raw documents in plain text, so file-system permissions are always sufficient
- C. They are used only for backups, so no production access model is needed
- D. Files are stored as vectors with metadata, so attribute, index or query based controls apply
Answer: D
Vector databases hold embeddings and metadata rather than the original text, so attribute-, index- or query-based controls are considered. They do hold metadata and are not backup tools. The access model differs from file-based controls.
Domain 1: AI Governance and Program Management
Q290 A dataset has missing mandatory fields and mismatched date formats. Which data quality dimensions are MOST affected?
- A. Completeness and consistency
- B. Timeliness and validity only
- C. Accuracy and availability only
- D. Integrity and confidentiality
Answer: A
Missing mandatory fields affect completeness and mixed date formats affect consistency. Timeliness and validity could also matter but are not the issue described. Confidentiality and integrity are security properties, not these quality dimensions.