CertArea
All Snowflake exams

Free ADS-B01 practice test: 30 questions with answers

30 original questions for SnowPro Specialty: AI Governance & Security (beta), drawn from every topic area. Answer each one to see whether you got it right and why. No sign-up.

Free questions
30
Full bank
112 questions

ADS-B01 exam guide: topics, cost and format

All 30 ADS-B01 questions with answers

The same questions as the test above, grouped by exam topic. Open any question to see the answer and explanation.

Cortex AI risk assessment (5 questions)

  1. Question 1

    A bank plans to let analysts ask a Cortex-powered assistant about customer complaint records, which contain names and account numbers. Before anything goes live, which step belongs first in the risk assessment?

    1. ATune the prompt wording so the assistant gives shorter answers
    2. BMeasure how many concurrent analysts the assistant can serve
    3. CIdentify which complaint fields are sensitive and who may see them
    4. DCompare response quality across several language models in a sandbox
    Show answer and explanation
    Correct answer: C

    Identifying the sensitive fields and who may view them defines the risks and the controls that are needed, so it comes first. Prompt tuning, capacity testing and model comparison are useful later but do not establish the risk baseline.

  2. Question 2

    A team builds a chatbot that answers questions using internal wiki pages. A tester places a crafted instruction inside one wiki page, and the chatbot then reveals content from pages it would normally not cite. Which risk category best describes this?

    1. AIndirect prompt injection carried by untrusted page content
    2. BModel hallucination caused by missing grounding documents
    3. CDenial of service caused by a flood of concurrent chatbot queries
    4. DSchema drift caused by renamed columns in a source table
    Show answer and explanation
    Correct answer: A

    Instructions hidden in content that the model ingests are indirect prompt injection. Hallucination means invented content, denial of service is an availability risk and schema drift is a pipeline issue, so none of those fit the scenario.

  3. Question 3

    Which statement best describes an entry in a risk register for a generative AI feature?

    1. AIt stores a complete copy of the training data so the feature can be retrained quickly
    2. BIt lists every model version the vendor has released since the feature first became available
    3. CIt records each risk with its rating, existing controls, residual level and an owner
    4. DIt replaces the security review that is normally required before a feature is approved
    Show answer and explanation
    Correct answer: C

    A risk register tracks each identified risk with its rating, the controls in place, the residual level and an accountable owner. The other options describe a model inventory, a training-data store and a review process, which are different artefacts.

  4. Question 4

    A product owner wants to send free-text customer feedback to a language model each week to find common themes. Some feedback includes health details. What is the most appropriate first control?

    1. AInstruct the model in the prompt to ignore any health details in its answer
    2. BStore every prompt and completion in a table so reviewers can audit later
    3. CLimit the trial to the ten largest customers by revenue for one quarter
    4. DMask or remove health details before the text is sent to the model
    Show answer and explanation
    Correct answer: D

    Sensitive content should be removed or masked before it leaves the controlled data boundary, because a prompt instruction does not stop the data from being sent or stored. Logging after the fact and limiting the trial size do not reduce the exposure of health details.

  5. Question 5

    Which two risks should a risk assessment for a generative AI feature normally cover? (Choose two.)

    1. ASensitive data in prompts being retained or exposed beyond approved limits
    2. BPricing changes announced for the warehouse compute tiers next year
    3. CUsers treating generated answers as verified facts without any review
    4. DInconsistent column names for the same dimension across environments
    5. EIdle time between scheduled jobs on a virtual warehouse
    Show answer and explanation
    Correct answer: A, C

    Exposure of sensitive data through prompts and over-reliance on unverified output are core generative AI risks. Pricing, naming consistency and warehouse idle time matter to operations and cost, but they are not AI-specific risk categories.

AI access control and agent identity (5 questions)

  1. Question 6

    A data engineer sets up a scheduled job that calls Cortex functions. The job currently runs under the engineer's personal user, which has broad privileges. The team wants the job to keep running if the engineer changes roles. Which approach is best?

    1. AKeep the engineer's personal user with its broad existing privileges and a password that never expires
    2. BCreate a dedicated service user with only the needed privileges, authenticated by key pair
    3. CGrant the job's role ACCOUNTADMIN so that no permission errors occur during scheduled runs
    4. DShare the engineer's password with the team and rotate it every quarter
    Show answer and explanation
    Correct answer: B

    A dedicated service identity with least-privilege grants and key-pair authentication survives staff changes and limits what a leaked credential can reach. The other options keep a personal identity, grant excessive rights or spread a shared password.

  2. Question 7

    Which statement best describes how Snowflake role-based access control delivers privileges to users?

    1. APrivileges go to roles, roles are granted to users or other roles, and active roles set access
    2. BPrivileges attach to databases, so any user who connects to that database automatically receives them
    3. CPrivileges attach to warehouses, which pass them on to whoever runs queries on those warehouses
    4. DPrivileges attach directly to each user, and roles only group users for reporting
    Show answer and explanation
    Correct answer: A

    Snowflake uses a role hierarchy: privileges are granted to roles, roles are granted to users or to other roles, and the roles active in a session determine effective access. The other options describe models that RBAC does not use.

  3. Question 8

    A chatbot queries a support articles table, and some articles are restricted to premium customers. The team needs each user to receive only the rows they are allowed to see, whatever the query path. Which design fits best?

    1. AAdd an instruction to the system prompt telling the model to skip restricted articles
    2. BCopy the restricted articles into a second table and hide the link to it in the interface
    3. CFilter results in the front-end application after the query has returned all rows
    4. DAttach a row access policy that checks each row's tier against the caller's role
    Show answer and explanation
    Correct answer: D

    A row access policy enforces the rule inside the database, so every query path, including AI features, sees only the rows the caller is permitted to see. Prompt instructions, hidden links and front-end filters are not enforced controls and can be bypassed.

  4. Question 9

    Which two actions best limit what an AI agent's service identity can reach? (Choose two.)

    1. AGrant the agent's role SELECT only on approved secure views rather than on base tables
    2. BRun the agent under a shared administrator login that already holds every privilege
    3. CMake the agent's role the owner of the sensitive base tables so it can manage them
    4. DReuse a developer's personal role because it already has the grants the agent needs
    5. ELimit the service user to approved warehouses and attach a network policy to it
    Show answer and explanation
    Correct answer: A, E

    Scoping grants to approved views and restricting the service user's warehouses and network access shrink the data and entry points the agent can reach. Shared admin logins, ownership of base tables and reused personal roles all widen what the agent can reach.

  5. Question 10

    A stored procedure that reads a sensitive table is created with the default owner's rights. A user who has no direct SELECT on the table calls the procedure. What is the expected outcome?

    1. AThe call fails, because owner's rights procedures never run statements with the owner's privileges
    2. BThe call can succeed, because statements run with the owner's privileges, which can read the table
    3. CThe call returns no rows, because the caller's missing grant hides the table from the procedure
    4. DThe call bypasses every masking policy on the table, because owner's rights ignore data protection
    Show answer and explanation
    Correct answer: B

    With owner's rights, the procedure's SQL runs with the owner's privileges, so the caller can get results without direct table access, which is why procedure ownership must be reviewed. The other options misstate how owner's rights work, including the claim that masking is bypassed.

Cortex AI Guardrails and Cortex Guard (5 questions)

  1. Question 11

    Which statement best describes the purpose of a guardrail on a generative AI call?

    1. AIt limits which users and roles may call the model by enforcing role-based privileges on the function
    2. BIt caches earlier responses so that a repeated prompt is served without a new model call
    3. CIt screens inputs or outputs against a safety policy, such as unsafe content, before the response is used
    4. DIt shortens long prompts so that more of the model's context window is left for the answer
    Show answer and explanation
    Correct answer: C

    Guardrails check content against a policy, for example unsafe material, and can block or replace what fails. Access control, caching and prompt shortening are separate mechanisms that do not filter content.

  2. Question 12

    A customer-facing assistant built on a COMPLETE call sometimes returns violent or hateful text when users probe it. The team wants unsafe responses filtered before they reach the app. Which Cortex option addresses this most directly?

    1. AA larger context window, so the model has more instructions to follow
    2. BCortex Guard, enabled on the call, to filter unsafe or harmful model responses
    3. CA lower temperature setting, which makes the model's answers more deterministic
    4. DAccess history reviews, which record each query and the objects it touched on the account
    Show answer and explanation
    Correct answer: B

    Cortex Guard screens model output for unsafe or harmful content when it is enabled on the call. A larger context window and a lower temperature do not filter harmful text, and access history is an audit record rather than a content control.

  3. Question 13

    A team protects pricing data in a Cortex chatbot with a system prompt that says never to reveal internal pricing. Every role that can use the chatbot can also read the pricing table. What is the main weakness?

    1. APrompt wording is not access control, so the table needs enforced restrictions
    2. BSnowflake ignores system prompts and applies only role-based policies to the chatbot's SQL output
    3. CThe system prompt is read only during the first message of a session and is then dropped
    4. DPricing values stored as numbers cannot be described by a language model in plain text
    Show answer and explanation
    Correct answer: A

    A prompt can be overridden by crafted user input, so sensitive rows need enforced protection such as row access or masking policies on the table itself. The other options describe behaviour that does not exist or that is irrelevant to protecting the table.

  4. Question 14

    Which two practices make a generative AI feature that answers questions over internal documents safer? (Choose two.)

    1. AEnable a content filter on model responses for unsafe or harmful text
    2. BRely on the model's default behaviour and skip further review because the model is vendor-managed
    3. CValidate and limit the format and length of user input before building the prompt
    4. DGive the feature's role full administrative privileges so that errors are avoided
    5. EPublish every blocked response to a public page so that users can see what was rejected
    Show answer and explanation
    Correct answer: A, C

    Filtering model output for unsafe content and validating user input before it is placed in a prompt are layered controls that sit alongside access controls. Relying on vendor defaults, granting admin rights and publishing blocked responses all weaken the design.

  5. Question 15

    Which description best distinguishes input validation from output filtering in a generative AI pipeline?

    1. AInput validation edits the model's weights, while output filtering changes the cached results of earlier queries
    2. BInput validation runs after the answer is displayed to the user, while output filtering runs before any text reaches the model
    3. CBoth are the same check, and they differ only in which warehouse executes the code that runs them
    4. DInput validation checks the prompt before the model runs, while output filtering checks the model's response before it is shown
    Show answer and explanation
    Correct answer: D

    Input validation screens what goes into a prompt, and output filtering screens what the model returns before users see it. The reversed order in one option and the claims about weights and warehouses describe differences that do not exist.

Horizon Catalog governance and observability (5 questions)

  1. Question 16

    A governance team wants every email-address column to be masked for analysts, including columns that are added to new tables later. Which approach fits best?

    1. ACreate a tag for email data, attach a masking policy to that tag, and apply the tag to each column
    2. BAsk developers to prefix every new email column name with a common label so analysts can filter it out
    3. CCreate a separate masking policy for each table and update it whenever a new table is added
    4. DTrust analysts to leave email columns out of their SELECT lists in their own queries
    Show answer and explanation
    Correct answer: A

    A masking policy attached to a tag applies to every column that carries the tag, so new columns are covered as soon as they are tagged. Per-table policies need constant upkeep, naming conventions are not enforced, and relying on analysts is not a control.

  2. Question 17

    Which Horizon Catalog capability shows how data flows from source tables through views and transformations into a downstream report?

    1. AData lineage, which maps upstream and downstream relationships between objects
    2. BAccess history, which lists the queries that read each object and the user who ran them
    3. CResource monitors, which cap the credits that warehouses can consume in a period
    4. DNetwork policies, which restrict the client IP addresses that may connect to the account
    Show answer and explanation
    Correct answer: A

    Data lineage traces the upstream sources and downstream consumers of objects, which answers how data flows. Access history records who read what, resource monitors control credit use and network policies control connections, so none of them show the flow.

  3. Question 18

    An auditor asks which users read a sensitive table during the last month, including reads through views that sit on top of it. Which source is the best starting point?

    1. AThe table comment field, which lists the teams that the owner has approved to read the table
    2. BThe ACCESS_HISTORY view in ACCOUNT_USAGE, which records objects accessed by each query and its user
    3. CThe query profile of the most recent run of each dashboard, which shows the execution steps
    4. DThe default role set on each schema, which shows which users were most recently granted access to it
    Show answer and explanation
    Correct answer: B

    ACCESS_HISTORY records the objects each query accessed, including base objects behind views, along with the user and role, which is the evidence an auditor needs. Comments document intent rather than actual use, query profiles cover single executions and default roles do not show reads.

  4. Question 19

    What does sensitive data classification in Horizon Catalog do?

    1. AIt encrypts flagged columns at rest with a customer-managed key that is held outside Snowflake
    2. BIt deletes columns that contain personal data once their retention period has passed
    3. CIt scans columns and assigns category tags, such as privacy categories, for use in policies
    4. DIt rewrites queries so that personal values are replaced by synthetic values at run time
    Show answer and explanation
    Correct answer: C

    Classification identifies likely personal or sensitive columns and tags them with categories that governance and tag-based policies can act on. Encryption, deletion and query rewriting are different controls that classification does not perform.

  5. Question 20

    A data steward wants to see, for each sensitive column, which tags are assigned and whether a masking policy is in effect, without reading any row values. Which approach fits best?

    1. AExport every sensitive table to a spreadsheet and check the masking by reading cells one by one
    2. BUse the tag and policy metadata for each column, which shows assignments without reading rows
    3. CQuery the first hundred rows of each sensitive table and inspect the values by eye
    4. DAsk each data owner to confirm verbally, in a meeting, that their sensitive columns are masked correctly
    Show answer and explanation
    Correct answer: B

    Tag and policy metadata show which tags are assigned and which masking policies are attached, without exposing row values. Sampling rows and spreadsheet exports either expose data or are slow and incomplete, and verbal confirmation is not evidence.

Data residency and compliance architecture (5 questions)

  1. Question 21

    A European insurer must keep its customer data inside the EU, but its disaster recovery plan calls for a copy in another region. Which approach best meets both needs?

    1. AReplicate only to a second region inside the EU that the compliance team has approved for storage
    2. BReplicate the database to a region on another continent, since that offers the most distance from a disaster
    3. CStore the data in an external stage on a non-EU cloud and copy it back to Snowflake when needed
    4. DKeep one account in a single non-EU region, since the plan requires no copy inside the EU at all
    Show answer and explanation
    Correct answer: A

    Replication must stay within regions that the residency rules allow, so a second EU region satisfies both the storage requirement and recovery planning. A copy on another continent, an external stage outside the approved boundary and a single non-EU region all break the stated constraint.

  2. Question 22

    Which statement about data residency in Snowflake is accurate?

    1. AClient location decides where stored data lives, so VPN settings determine the storage region
    2. BSnowflake balances storage across regions automatically to match the least busy cluster
    3. CData stays in the account's chosen region unless replication or another explicit feature moves it
    4. DChanging an account's region is a single admin toggle that keeps all existing data in place
    Show answer and explanation
    Correct answer: C

    A Snowflake account's data is stored in the region selected when the account was created, and it moves only through explicit actions such as replication. The other options describe storage behaviour that does not exist, such as client-driven storage location, automatic regional balancing and one-click region changes.

  3. Question 23

    A healthcare provider wants its own encryption key to be part of the protection of data stored in Snowflake, so that revoking that key can cut off access to the data. Which feature fits this requirement?

    1. APrivate connectivity through a cloud provider's private network endpoint
    2. BDynamic data masking applied to columns that hold identifiers
    3. CA network policy that allows connections only from approved client IP address ranges
    4. DTri-Secret Secure, which combines a customer-managed key with Snowflake's key
    Show answer and explanation
    Correct answer: D

    Tri-Secret Secure combines a customer-managed key with Snowflake's key to protect data at rest, so revoking the customer key can block access. Private connectivity, dynamic masking and network policies control network paths or query results, not the encryption key.

  4. Question 24

    Which two items should a compliance architect verify before promising that a Snowflake deployment meets a specific regulatory requirement? (Choose two.)

    1. AWhether the account's edition and the security features the requirement needs are actually enabled
    2. BWhether the warehouse size is large enough for the nightly loads to finish inside the batch window
    3. CWhether the account region and any cross-region features match the approved data residency rules
    4. DWhether the BI team's dashboards load quickly enough for the weekly management review
    5. EWhether the analytics project has been given a name that follows the internal naming convention
    Show answer and explanation
    Correct answer: A, C

    Compliance depends on the edition and controls the account actually has, and on where data is stored and processed. Warehouse sizing, dashboard speed and project naming have no bearing on whether a regulatory requirement is met.

  5. Question 25

    A regulator asks for documentation of where personal data is stored, how it is processed and who receives it. Which artefact best answers this request?

    1. AA warehouse inventory listing each warehouse name, its size, its auto-suspend and auto-resume settings
    2. BA record of processing activities listing data categories, locations, recipients and retention
    3. CA dashboard screenshot from the quarter's end showing query counts for the analytics team
    4. DThe vendor's marketing brochure describing platform performance and the uptime target
    Show answer and explanation
    Correct answer: B

    A record of processing activities is the documentation regulators typically expect, because it covers the categories, locations, recipients and retention of personal data. The other artefacts describe compute, marketing or usage and do not document how personal data is handled.

ACCOUNT_USAGE and Trust Center findings (5 questions)

  1. Question 26

    A security analyst runs a query on an ACCOUNT_USAGE view right after a suspicious login and finds no matching row. Which explanation is most likely?

    1. AThe ACCOUNT_USAGE views refresh continuously, so the login must have failed to record
    2. BThe view shows only logins made with the analyst's own role, so other users are hidden
    3. CLogin events are written to ACCOUNT_USAGE only after the user's session has been closed
    4. DACCOUNT_USAGE has built-in latency, so the newest events can take time to appear
    Show answer and explanation
    Correct answer: D

    ACCOUNT_USAGE is populated with a delay, so very recent events can be missing for a while. The other options describe behaviour that the view does not have.

  2. Question 27

    A governance analyst's role cannot see any rows in ACCOUNT_USAGE views. The analyst does not need ACCOUNTADMIN. Which grant is the usual fix?

    1. AGrant the role the ability to create warehouses, which unlocks the ACCOUNT_USAGE schema
    2. BSet the role's default warehouse to a larger size so that the views return more rows
    3. CCreate a view named ACCOUNT_USAGE inside the analyst's own database as a bridge
    4. DGrant imported privileges on the shared SNOWFLAKE database to that role
    Show answer and explanation
    Correct answer: D

    Read access to the ACCOUNT_USAGE views comes through the shared SNOWFLAKE database, so granting imported privileges on it lets the role query them. Warehouse creation, warehouse size and a local view with a matching name do not grant access to the shared data.

  3. Question 28

    What is the Snowflake Trust Center used for?

    1. ABrowsing third-party security apps that can be installed from a marketplace into the account
    2. BCertifying an organisation's compliance audit on Snowflake's behalf for regulators to accept
    3. CReviewing security scanner findings for an account, each with its status and recommended remediation
    4. DReporting the credits that each security-related feature consumed over the billing period
    Show answer and explanation
    Correct answer: C

    The Trust Center collects findings from security scanners for the account and shows the recommended remediation for each one. The other options describe audit certification, a marketplace and billing reports, none of which is the Trust Center's purpose.

  4. Question 29

    Which statement best describes which ACCOUNT_USAGE view to use when investigating failed login attempts by a service user over the past week?

    1. AWAREHOUSE_METERING_HISTORY, which shows the compute credits consumed by each individual warehouse
    2. BLOGIN_HISTORY, which records login attempts with the user and whether each one succeeded
    3. CGRANTS_TO_ROLES, which shows the privileges that each role holds at the present moment
    4. DSTORAGE_USAGE, which reports the bytes stored by the account on each day across all its databases
    Show answer and explanation
    Correct answer: B

    LOGIN_HISTORY records login attempts, including the user and whether each one succeeded, which is what a failed-login investigation needs. The other views report compute credits, current grants or storage, and none of them shows authentication outcomes.

  5. Question 30

    A cost analyst wants to see how many credits each team's warehouse consumed last quarter. Which ACCOUNT_USAGE view is the best starting point?

    1. AQUERY_HISTORY, which holds only the queries run in the current session, not earlier sessions
    2. BWAREHOUSE_METERING_HISTORY, which records the credits each warehouse consumed over time
    3. CGRANTS_TO_USERS, which shows the roles currently granted to each user in the account
    4. DSTORAGE_USAGE, which reports the bytes stored by the account on each day of the period
    Show answer and explanation
    Correct answer: B

    WAREHOUSE_METERING_HISTORY records credit consumption per warehouse over time, which is what chargeback needs. QUERY_HISTORY is account-wide rather than limited to one session, and the grants and storage views do not report compute credits.