What is the passing score for the Salesforce Certified AI Associate exam?

The passing score for the Salesforce Certified AI Associate exam is 65%.

How many questions are on the Salesforce Certified AI Associate exam?

The Salesforce Certified AI Associate exam consists of 40 multiple-choice questions.

Is the Salesforce Certified AI Associate exam hard?

The Salesforce Certified AI Associate is an entry-level exam and is generally considered easier than professional-level certifications. However, it requires a solid understanding of AI fundamentals, data concepts, and ethical AI practices within the Salesforce ecosystem to pass.

How much does the Salesforce Certified AI Associate exam cost?

The registration fee for the Salesforce Certified AI Associate exam is $75 USD, plus applicable taxes as required by local law. Retakes are currently offered for free.

Cisco

CompTIA

ISC

EC-Council

EXIN

Fortinet

ITIL

Juniper

Microsoft

VMware

Avaya

SAP

Google

NVIDIA

Salesforce

Amazon

Palo Alto Networks

ISC2

Splunk

ServiceNow

All Vendors

Salesforce Certified AI Associate Exam Questions
Salesforce Certified AI Associate

Updated On: 19-Apr-2026

Pass the Salesforce Certified AI Associate exam with community-verified questions and a free, built-in AI Tutor.

The Salesforce Certified AI Associate examination evaluates core competencies in implementing the Einstein Trust Layer, encompassing data privacy, prompt grounding, and model governance within the Salesforce ecosystem. Candidates must demonstrate technical proficiency in deploying generative AI, predictive modeling, and Large Language Models while mitigating toxic outputs and hallucinations through secure API integrations. Target audiences include Salesforce administrators, business analysts, and functional consultants tasked with configuring AI capabilities within CRM workflows. Proficiency requires mastery of data masking, retrieval-augmented generation fundamentals, bias detection, and ethical deployment strategies to ensure architectural compliance, data sovereignty, and security posture across distributed enterprise cloud environments.

AI Tutor: Every exam has a dedicated AI tutor. Don't just memorize—understand the why behind every correct answer.

12850
Students passed the Certified AI Associate Exam using this exam dump

96.4%
Average score during Real Exams at the Testing Centre

106
Exam Questions and Answers in this Certified AI Associate Prep

Salesforce
Salesforce-AI-Associate
Salesforce Certified AI Associate Exam

Total Questions: 102

Browse Free SALESFORCE-AI-ASSOCIATE Questions

Post your Comments and Discuss Salesforce Certified AI Associate exam dumps with other Community members:

Comments:

Name:

Certified AI Associate Exam Discussions & Posts

Community-Verified AI Explanation Commented on April 20, 2026
Question 1:
Question 1 asks which statements correctly describe language options when using Snowflake interfaces (select TWO).

A. Notebook cells can be written in Python. Correct. Snowflake Notebooks support Python as an executable cell type alongside SQL.

C. Notebook cells can be written in SQL. Correct. Notebooks also support SQL cells for declarative querying.

Why the others are not correct:

B. Databases can process queries using Scala. Not correct. Snowflake does not natively execute queries in Scala; Scala is available via Snowpark for application code, not as a direct query language.

D. Worksheets can only be written in SQL. Not correct. Snowsight worksheets are SQL-based but can also use Snowflake Scripting (procedural code), so they’re not restricted to SQL alone.

E. Dashboards can be written in JavaScript. Not correct. Dashboards are SQL-backed visualizations in Snowsight; JavaScript isn’t the authoring language there.

Key takeaway: A and C are the correct options because Snowflake Notebooks support both SQL and Python for notebook cells.
Innisfil, Canada

Community-Verified AI Explanation Commented on April 20, 2026
Question 12:

Correct answer: Test a flow with rollback

Why: Flow Designer supports:

- Call a subflow from a flow - Use Delegated Developer - Run a flow from a MetricBase Trigger The ability to test a flow with rollback is a capability of the Automated Test Framework (ATF), not Flow Designer.
Bengaluru, India

Community-Verified AI Explanation Commented on April 20, 2026
Question 7:

Answer: A — Association Response is a Management Frame.

Why:

- 802.11 frames are categorized into Data Frame, Management Frame, and Control Frame. - The Association Request/Association Response are part of the Management Frame used during the initial association process between a station and an AP. - The AP sends the Association Response in reply to the Association Request, indicating status and assigning an AID.
If you want, I can explain what’s included in the Association Response (e.g., status code, AID).
Harare, Zimbabwe

Community-Verified AI Explanation Commented on April 20, 2026
Question 21:

Answer: Utility (D)

Explanation:

In ITIL 4, value from a service comes from two properties: utility and warranty.

Utility = fitness for purpose; it asks whether the service provides the right functionalities and outcomes to meet customer needs.

Warranty = fitness for use; it covers availability, capacity, continuity, and security.

So to determine if a service is fit for purpose, assess its utility—does it deliver the required features and outcomes for the user?

Norwich, United Kingdom

Sumit Chaturvedi Commented on April 20, 2026
Is CAD Exam Have same questions every time or it changes on each attempts or changes few but not all, Please Suggest if some one have this info
Singapore, Singapore

Community-Verified AI Explanation Commented on April 20, 2026
Question 1:

Correct answer: Databricks web application (Option C)

Why:

- In the classic Databricks architecture, the control plane hosts the managed services that handle UI, authentication, and orchestration (the Databricks web application and related APIs). - The data plane runs the actual compute resources (the driver node and worker nodes) and storage interfaces like DBFS. - Therefore, the web application is hosted completely in the control plane, while the compute components (driver/worker) are in the data plane.
Pune, India

Community-Verified AI Explanation Commented on April 19, 2026
Question 77:

Answer: Billing Administrator

Why: Purchasing licenses is a licensing/billing action. The Billing Administrator role has the minimum rights to manage billing/subscriptions and license purchases for the subscription. The other roles are broader or unrelated to licensing:

- Global Administrator: full tenant-wide access (not least privilege). - Permissions Management Administrator: controls permissions management, not licensing. - User Access Administrator: manages access to resources, not licensing.

How to assign (brief): In the Azure portal, go to the subscription Sub1 ? Access control (IAM) ? Add role assignment ? select Billing Administrator ? assign to User1.

Rueil-Malmaison, France

Community-Verified AI Explanation Commented on April 19, 2026
Question 97:
Question 97 explains you need a CA policy for App1 that allows only Windows devices and only compliant devices.

Correct settings:

- Condition: Device platforms -> Windows - Grant: Require device to be compliant

Why:

- Device platforms restrict which devices can even be evaluated (Windows in this case). - The grant control enforces that the device is compliant (Intune/compliance policy)
Tübingen, Germany

Community-Verified AI Explanation Commented on April 19, 2026
Question 91:

Answer: The maximum number of user risk policies you can configure is 1 (Option A).

Why: In Azure AD Identity Protection, there are two policy types that use risk signals: User risk policies and Sign-in risk policies. The limit for User risk policies is one per tenant. You configure this single policy to define actions (e.g., require MFA, block) based on detected user risk levels. If you need additional controls, use Sign-in risk policies (a separate policy type) in addition to the one user risk policy.

Tübingen, Germany

Community-Verified AI Explanation Commented on April 18, 2026
Question 30:

Correct answer: B - Deploy the monitoring pod in a DaemonSet object.

Explanation:

A DaemonSet ensures exactly one pod runs on every node (and on new nodes added by cluster autoscaler). As pods are added/removed with nodes, the DaemonSet keeps per-node coverage for metrics collection.

Why the others are not suitable:

- StatefulSet: for stable identities and persistent storage, not per-node coverage. - Deployment: manages replicas, but there’s no guarantee a pod will exist on every node. - Cluster Initializer: irrelevant for workload placement (not used for this scenario).
Tigre, Argentina

Community-Verified AI Explanation Commented on April 18, 2026
Question 5:

Correct answer: A. Use gcloud config configurations describe to review the output.

Explanation:

gcloud supports multiple named configurations. Each configuration stores settings like project, zone, and region.

To view a specific inactive configuration without switching to it, use:

- gcloud config configurations describe <CONFIG_NAME>

This directly shows the settings for that configuration.

The other options involve activating the configuration or using kubectl contexts, which are more steps or unrelated to inspecting the gcloud configuration itself.

Tigre, Argentina

Community-Verified AI Explanation Commented on April 18, 2026
Question 76:

Correct answer: Orchestration

Why: To use Workflow and Client Software Distribution (CSD) for automatic software installation on devices, you need the separate license for Orchestration. ITSM and Asset Management cover service/support processes and license tracking, not cross-system automation. Discovery is about finding devices, not deploying software, and Deployment Management (a separate capability) is not the feature that enables CSD installations. Orchestration enables remote actions and software deployment across endpoints via workflows.

Johannesburg, South Africa

Community-Verified AI Explanation Commented on April 18, 2026
Question 7:
Não consigo ver o conteúdo do [Image] de Questão 7 nem as opções. Pode colar o trecho JSON apresentado ou descrever as afirmações e as opções disponíveis? A partir disso eu marco qual é a correta e explico o raciocínio.
Enquanto isso, aqui vão dicas úteis sobre JSON para Questões de hot area:

Arrays JSON:

- Representados com [ ]. - São uma lista ordenada; a ordem importa. - Elementos podem ser de qualquer tipo JSON (número, string, objeto, array, boolean, null). - Acesso típico: indexação, por exemplo, arr[0].

Objetos JSON:

- Representados com { }. - São coleções de pares chave-valor; as chaves são strings. - Em JSON, a ordem de pares não é garantida. - Acesso típico: por chave, por exemplo, obj["name"] ou obj.name (quando disponível).

Dicas úteis para as questões:

- Identifique se a afirmação descreve uma estrutura de array (ordem, acesso por índice) ou de objeto (pares chave-valor, acesso por chave). - Verifique se as regras de sintaxe estão corretas: strings entre aspas duplas, tipos permitidos, etc. - Se houver comparação entre arrays e objetos, confirme quem é ordenado vs não ordenado, e como se acessa cada um.
Envie o conteúdo (json e opções) e eu sigo com a explicação objetiva da Questão 7.
Rio De Janeiro, Brazil

Community-Verified AI Explanation Commented on April 18, 2026
Question 3:

Correct answer: ETL (Extract, Transform, Load).

What ETL means: A data pipeline that collects data from multiple sources, applies business rules to transform the data, and then loads it into a destination data store (like a data warehouse).

The three steps:

- Extract: Pull data from various source systems. - Transform: Clean, normalize, deduplicate, and combine data according to business rules. - Load: Write the transformed data into the target store (data warehouse or data lake).

Why it’s used: Improves data quality and consistency, enables analytics-ready data, and centralizes data for reporting.

In DP-900 context: ETL is a common pattern for preparing data for analytics. Tools like Azure Data Factory orchestrate ETL workflows. Note that some modern setups use ELT (Extract, Load, Transform) when the transformation happens inside the target database, but the question specifically defines ETL order.

Rio De Janeiro, Brazil

Community-Verified AI Explanation Commented on April 18, 2026
Question 4:

Correct answer: Observed

Why: The Observed mode uses real-time feedback on each pool member’s performance (latency/throughput) to distribute traffic. This helps keep the load balanced across all six servers when client interactions vary. Round Robin distributes evenly by order but doesn’t account for current load; Ratio and Priority use static or hierarchical rules and may not adapt to performance changes. Observed adapts to maintain a relatively even load.

Seattle, United States

Community-Verified AI Explanation Commented on April 18, 2026
Question 3:
I can’t view the image in Question 3, but the listed answer is A. Here’s how to approach this hotspot question conceptually.

Objective: identify the host that originated the infection and classify the other hosts as clean or infected based on logs.

Evidence to review:

- firewall logs: look for anomalous outbound connections, unusual ports, or traffic from a single host to multiple others. - host logs (security/EDR): look for first signs of malware, suspicious process creation, PowerShell/WMIC usage, new user accounts, or credential access.

Determine the origin:

- Build a timeline from earliest to latest events. - The origin is typically the host with the earliest suspicious activity or the source of initial lateral movement/inbound connections to others.

Classify remaining hosts:

- Infected: logs show evidence of compromise linked to the origin (same IOC, remote execution, or propagation indicators). - Clean: no evidence of compromise or propagation from the origin.

If you can describe what you see in the logs or on which host the early activity appears, I can map it onto the options and explain why A is the correct choice.

Grand Prairie, United States

Community-Verified AI Explanation Commented on April 18, 2026
Question 16:

Correct answer: B (UI Pages)

Explanation:

The History section of the Application Navigator shows items you’ve recently opened that are user data views, such as:

- Records (a specific table record) - Lists (a list view of a table) - Forms (the form view for a record)

UI Pages are UI configuration assets, not typical data views you’d open as a user, so they do not appear in the History. To access UI Pages, use the path under System UI (e.g., System UI > UI Pages) rather than the History.

United States

Community-Verified AI Explanation Commented on April 18, 2026
Question 44:
Question 44 explains protecting an S3 bucket from accidental deletion and asks for two correct steps.

Correct answers: A and B

- A. Enable versioning on the S3 bucket. Keeps previous versions of objects, so you can recover if you delete or overwrite by mistake. - B. Enable MFA Delete on the S3 bucket. Requires MFA to permanently delete object versions or delete markers, adding a strong safeguard.
Why these work:

Versioning provides data recovery from deletions or overwrites.

MFA Delete adds an extra layer, preventing permanent deletions even if someone has access to the AWS account.

Why the others are not correct:

C. Create a bucket policy alone doesn’t prevent d
United States

Community-Verified AI Explanation Commented on April 18, 2026
Question 41:
Question 41: You want to improve performance with the least operational overhead for SaaS-to-S3 data ingestion and completion notification.

Answer: B — Create an Amazon AppFlow flow to transfer data between each SaaS source and the S3 bucket, and configure an S3 event notification to publish to an SNS topic when the upload completes.

Why this is correct:

AppFlow is a managed service that orchestrates data transfers from multiple SaaS sources to S3 with automatic scaling, reducing the need to manage EC2-based logic or custom connectors.

Coupling AppFlow with an S3 event notification to SNS provides completion notifications without building and maintaining custom pipe
European Union

SuPeRLeo Commented on April 18, 2026
QUESTION: 9 is obviously D.. NOT B etc
Nixa, United States

babubhatta634 Commented on April 18, 2026
what is diffrence use of azure service bus queue vs azure service bus topic

In Azure Service Bus, choose based on the communication pattern: point-to-point (queue) vs publish-subscribe (topic).

Queue

- Purpose: point-to-point messaging. A message is delivered to only a single consumer. - Pattern: competing consumers (scale by adding more receivers, but each message goes to one). - Use cases: task queues, order processing, background work where only one worker should handle each message. - Features: dead-letter, TTL, retries, potential ordering with sessions.

Topic

- Purpose: publish-subscribe messaging. A message can be delivered to multiple independent receivers. - Pattern: fan-out. One message is copied to every matching subscription. - Use cases: event distribution, notifying multiple apps/services (billing, analytics, CRM) from a single event. - Features: subscriptions with filters/rules, per-subscription dead-lettering, independent TTL and delivery settings per subscription.

Quick decision guide:

- Use a Queue when you need one consumer process each message. - Use a Topic when you need to broadcast the same message to multiple apps/services (with optional filtering per subscriber).
Kolkata, India

Community-Verified AI Explanation Commented on April 17, 2026
Question 12:

Correct answer: Hosts file

Why: The hosts file is a local override for DNS. If a wrong entry exists for the URL in a single user’s hosts file, that user will resolve the URL to an incorrect IP, while others using the real DNS can reach the site.

Why the other options are less likely:

- Self-signed certificate: affects TLS trust, not DNS resolution. - Nameserver record: would typically affect many users if the DNS server is misconfigured. - IP helper: unrelated to hostname-to-IP resolution.

Quick checks:

- On Windows, inspect C:\Windows\System32\drivers\etc\hosts and look for the URL mapping; remove/adjust if present. - On macOS/Linux, check /etc/hosts. - Test with nslookup <URL> or ping <URL> to confirm what IP is being resolved. - Clear cache if needed (ipconfig /flushdns on Windows).
Passaic, United States

Community-Verified AI Explanation Commented on April 17, 2026
Question 3:

Question 3 asks if subscribing to Premier (to get a Microsoft architectural/design assessment) meets the goal of getting the assessment while keeping costs to a minimum.

Basic does not support architectural assessments. Premier does include an architectural/design review, but it is the most expensive option.

Therefore, while Premier enables the requested service, it does not satisfy the cost-minimization requirement.

Answer: No
Auckland, New Zealand

kuwarsahab17 Commented on April 17, 2026
Question 31:

Correct answer: D — service proxy and control plane.

Why:

In a service mesh, the data plane is made of service proxies (often as sidecars) that handle actual inter-service traffic, including routing, load balancing, mTLS, and telemetry.

The control plane manages configuration, policies, service discovery, and distributes config to the proxies.

The other options describe observability (A), Kubernetes scheduling concerns (B), or use an incorrect term (C: “runtime plane” isn’t a standard component).

Bengaluru, India

Community-Verified AI Explanation Commented on April 17, 2026
Question 18:

Correct answer: StatefulSet

Why:

- A headless service is created by setting clusterIP: None. It provides direct DNS entries for each pod (e.g., myservice-0, myservice-1), enabling stable network identities. - StatefulSets require stable, unique network identities and ordered startup/shutdown. A headless service supports direct pod addressing without load balancing. - Other workloads like Deployment, DaemonSet, and CronJob typically use a regular service with a clusterIP to load balance traffic across pods, not required for per-pod DNS.
Bengaluru, India

Community-Verified AI Explanation Commented on April 17, 2026
Position and job opening is a Master Detail relationship. OWD for Position is Private. Someone else wants to edit Job Openings. How to do it?

Answer: Create Sharing rule in Position

Why: In a Master-Detail setup, child records (Job Openings) inherit access from the parent (Position). If Position’s OWD is Private, you must grant access to the Position records to the user (via a sharing rule or manual sharing) to allow edits to the related Job Openings. A sharing rule on Position automates this for groups/roles; apex-based sharing isn’t needed.

São Paulo, Brazil

Community-Verified AI Explanation Commented on April 17, 2026
Question 10:
Question 10 answer: A (PKI)
Why: X.509 is the standard certificate format used by a Public Key Infrastructure (PKI). These certificates bind a public key to an identity and are widely used to authenticate and encrypt communications (e.g., TLS/HTTPS).
Other options:

VLAN tagging: unrelated to certificates.

LDAP: directory service protocol, not certificate format.

MFA: authentication method, not certificate framework.

Passaic, United States

Community-Verified AI Explanation Commented on April 17, 2026
Question 91:

Answer: B (False)

Explanation:

AWS PrivateLink creates private endpoints inside your VPC to Snowflake’s service endpoints, enabling private connectivity from your AWS resources to Snowflake.

It does not directly provide a secure connection from your on-premises data center to Snowflake. To reach Snowflake from on-prem, you’d typically connect your on-prem network to your AWS VPC (via a VPN or AWS Direct Connect) and then use PrivateLink from that VPC to Snowflake.

Hyderabad, India

Community-Verified AI Explanation Commented on April 17, 2026
Question 143:
Correct answer: C. Meeting stakeholder expectations
Why this is correct:

The Design & Transition activity focuses on designing new/changed services to deliver value and to be transitioned smoothly into live operation. A primary goal is to ensure the service meets what stakeholders expect from it.

Why the others are less fitting:

A) Understanding the organization's vision: This is strategic guidance, not the core focus of design and transition.

B) Understanding stakeholder needs: Important, but the main concern is delivering to stakeholders’ expectations, i.e., the outcomes.

D) Ensuring service components are available: This relates more to availability and operations, not the central design/transition concern.

Chennai, India

Community-Verified AI Explanation Commented on April 17, 2026
Question 123:

Correct answer: D: Continual improvement.

Why: The Continual Improvement Practice focuses on implementing and accelerating improvements across services and practices. It explicitly embraces modern approaches like Lean, Agile, and DevOps to enable more frequent, value-driven change.

Why the others aren’t correct:

- Service desk is a function for handling incidents and requests, not a change-facilitating approach. - Monitoring and event management deals with detecting and managing events, not driving rapid organizational change. - Service level management focuses on defining and agreeing service levels, not on enabling faster change.

Quick tip: When a question mentions Lean, Agile, or DevOps, think of continual improvement as the ITIL 4 practice designed to accelerate and improve change delivery.

Chennai, India

Community-Verified AI Explanation Commented on April 17, 2026
Question 15:

Correct answer: Yes (A)

Why: The goal is to build, test, and deploy predictive analytics. Azure Machine Learning Studio provides an end-to-end ML lifecycle: data preparation, model training, evaluation (testing), and deployment to endpoints. It supports pipelines, versioning, and deployment of predictive analytics, aligning with the required workflow.

Karimnagar, India

Community-Verified AI Explanation Commented on April 17, 2026
Question 38:

Correct answer: NFV

Explanation:

- NFV (Network Functions Virtualization) is the framework that allows network functions like routing to run as software on standard hardware inside a hypervisor as virtual network functions (VNFs). - This is opposed to: - VPC (virtual private cloud) — a cloud construct, not a routing capability inside a hypervisor. - IaaS — a service model, not a specific network function. - Firewall — a function, but not the virtualization framework that enables routing on the hypervisor.
Harlow, United Kingdom

Anon Commented on April 17, 2026
Question 74 the answer should be option C - users and mail enabled security groups, instead the answer shows option D and yet explains for option C, lol
Pune, India

Community-Verified AI Explanation Commented on April 16, 2026
Question 41:

Correct answers: IntegrationLatency and Latency.

Why:

- IntegrationLatency measures the time API Gateway spends calling the backend integration (your Lambda function) and waiting for its response. This helps determine if the delay is inside the Lambda integration. - Latency measures the total time from when API Gateway receives the request to when it sends the response back to the client. This helps identify if the timeout is due to the entire API Gateway path (including any network or gateway processing) rather than just the Lambda call.

Why not CacheHitCount/CacheMissCount: those relate to API Gateway caching. If you’re not using caching (or the problem isn’t caching-related), they won’t help diagnose a timeout.

Practical tip: in CloudWatch, compare Latency and IntegrationLatency for the failing method/resource. If Latency is high but IntegrationLatency is low, the delay is outside the Lambda integration (gateway/network). If IntegrationLatency is high, the Lambda/backend is the bottleneck. Also review CloudWatch Logs for API Gateway and Lambda for additional clues.

Melbourne, Australia

Community-Verified AI Explanation Commented on April 16, 2026
Question 61:

Correct answers: A and D

Why:

- A: to enable intra-enterprise communication — Private addresses can be used inside an organization for internal hosts to talk to each other without needing globally unique addresses. - D: to conserve global address space — Private IPv4 addresses are not routable on the Internet, so using them preserves the limited public IPv4 addresses.

Why the others aren’t primary reasons:

- B: to implement NAT — NAT uses private addresses, but this is a consequence, not the main reason to use private space. - C: to connect applications — Not specific to the purpose of private IP space. - E: to manage routing overhead — Not a primary objective of private addressing.
London, United Kingdom

Community-Verified AI Explanation Commented on April 16, 2026
Question 59:
Correct options: dual-homed branches and dynamic routing.

Dual-homed branches: Each branch has two independent WAN paths (often via different providers). This provides redundancy (reliability) and, with routing choices, helps distribute traffic (scalability) if one link or provider fails.

Dynamic routing: Uses routing protocols (e.g., OSPF, EIGRP, BGP) to automatically learn and adjust routes. It scales well as the network grows and improves convergence and resilience after topology changes.

Why the others aren’t correct:

Asynchronous routing: not a standard WAN design term here.

Single-homed branches: single path, low reliability.

Static routing: Good for small setups but not scalable or resilient to changes.

London, United Kingdom

Community-Verified AI Explanation Commented on April 16, 2026
Question 57:
Question 57 asks: Which two goals/reasons to implement private IPv4 addressing on your network?

Correct answers:

- Conserve IPv4 address — using RFC 1918 private address space means internal hosts don’t consume public IPv4 addresses, often paired with NAT for Internet access. - Reduce the risk of a network security breach — private addresses are not publicly routable, so external hosts cannot directly reach internal devices; NAT adds an extra layer of obscurity.

Why the others aren’t correct:

- Comply with PCI regulations – not a direct requirement of private addressing. - Reduce the size of the forwarding table on network routers – private addressing doesn’t inherently shrink internal router FIB size. - Comply with local law – private addressing isn’t a legal requirement.
Key concept: Private IPv4 addresses (RFC 1918) are used inside networks to conserve public addresses and reduce exposure to the Internet, typically with NAT for outbound traffic.
London, United Kingdom

Community-Verified AI Explanation Commented on April 16, 2026
Question 2:

Correct answer: D. Use Vertex AI Agent Builder to create a custom AI agent.

Why this is correct:

The goal is to answer questions about papers and provide summarized insights without heavy coding.

Vertex AI Agent Builder enables creating custom AI agents with low-code/no-code capabilities that can process scientific documents and generate QA responses and summaries.

Why the other options are less suitable:

A: Gemini for Google Workspace is for collaborative document review, not for building a document-answering agent.

B: Vertex AI Search indexes papers for keyword searches but doesn’t directly answer questions or provide summaries.

C: Vertex AI AutoML trains classification models, which is not the same as building a QA/summarization agent.

Québec, Canada

Community-Verified AI Explanation Commented on April 16, 2026
Question 1:

Correct answer: D. Use Google Cloud databases and Vertex AI for the agent to get live data.

Why this is correct:

The agent must access real-time internal inventory data to check levels and adjust delivery schedules.

Google Cloud databases provide the live data store for current inventory.

Vertex AI offers the platform to build, deploy, and manage the AI agent and connect it to live data sources.

Why the other options are less suitable:

A: Building a custom API for every interaction adds unnecessary complexity and cost; a direct live-data connection is more scalable.

B: Pre-built chatbots usually don’t integrate with private, real-time internal systems required for inventory and scheduling updates.

C: Fine-tuning a model with sample data does not give live data access, so it cannot respond to current inventory or scheduling needs.

Québec, Canada

Community-Verified AI Explanation Commented on April 16, 2026
Employees located off-site must have access to company resources in order to complete their assigned tasks. These employees utilize a solution that allows remote access without interception concerns. Which of the following best describes this solution?

Answer: VPN

Explanation: A VPN provides secure, encrypted remote access to the corporate network, creating a tunnel that protects data from interception for off-site workers.

- Proxy server: does not inherently encrypt end-to-end. - NGFW: focuses on firewalling and inspection, not remote access. - Security zone: relates to segmentation, not remote access.
Johannesburg, South Africa

Community-Verified AI Explanation Commented on April 16, 2026
Question 50:

Correct answer: A. Purchase Provisioned Throughput for the custom model.

Why: When using a private/custom model in Amazon Bedrock, you must allocate capacity by purchasing provisioned throughput to guarantee predictable latency and scale for real-time inference. Bedrock serves the model through its own hosting, so deploying to a SageMaker endpoint or registering it in SageMaker Model Registry isn’t required for Bedrock usage. Access control is managed via Bedrock/IAM, but the provisioning step is the mandatory one to enable serving.

Bengaluru, India

Community-Verified AI Explanation Commented on April 16, 2026
Question 4:
Correct answer: B, C

Service Cloud focuses on customer service and support.

Knowledge: Used for a knowledge base of articles to assist agents and customers.

Entitlements: Define service levels/SLA for handling cases.

Campaigns and Opportunities/Quotes are typically part of Sales Cloud (marketing and sales functionalities, respectively).

Pune, India

Community-Verified AI Explanation Commented on April 16, 2026
Question 1:

Correct answer: D — Promote off-label use to a carefully selected patient population.

Why:

Conditional approval means confirmatory studies are still required. To minimize risk during this period, you should implement risk minimization actions that keep use within approved terms.

Labeling for appropriate populations ensures clinicians and patients know exactly who the product is intended for.

Educating patients and healthcare providers reduces misuse and misinterpretation of how to use the product.

Delaying product launch limits patient exposure to potential unknown risks until safety/efficacy are confirmed.

Promoting off-label use is least effective (and inappropriate): it increases risk, bypasses approved indications, and undermines regulatory commitments. Off-label use lacks the evidentiary basis and regulatory safeguards of approved indications.

Jinrongjie, China

openiyi2613 Commented on April 16, 2026
Question 1:

Correct answer: runC

Why: The Open Container Initiative (OCI) defines a runtime specification for spawning and managing containers. The native, reference implementation of this spec is runC (the OCI runtime). It runs containers directly on the host kernel using Linux namespaces, cgroups, etc.—this is what makes it the OCI-compliant native runtime.

Why the others are not considered the native OCI runtime:

- runV: an older VM-based runtime, not the standard OCI runtime. - kata-containers: VM-based isolation layered on top of OCI, not the native host-process runtime. - gvisor: a user-space kernel sandbox for containers, not an OCI runtime itself.
Lagos, Nigeria

Community-Verified AI Explanation Commented on April 16, 2026
tftp full form

TFTP full form: Trivial File Transfer Protocol.

Quick note: It’s a simplified file transfer protocol (uses UDP/port 69) often used in networking for loading IOS images or configurations to/from Cisco devices. This topic can appear in 200-301 when discussing device file transfers.

Mumbai, India

Community-Verified AI Explanation Commented on April 16, 2026
Question 51:

Correct answer: the Microsoft 365 connector (option C).

Why:

To investigate threats using data in the unified audit log of Microsoft Defender for Cloud Apps, you must first connect Microsoft 365 to Defender for Cloud Apps.

The Microsoft 365 App Connector enables Defender for Cloud Apps to ingest data (including audit logs) from Microsoft 365 so you can correlate and investigate threats.

What to do (high level):

In Microsoft 365 Defender portal: Settings > Cloud Apps > App Connectors.

Click +Connect an app and select Microsoft 365.

Choose the necessary components, click Connect, and follow links to complete the setup.

Verify the connector shows as Connected.

Why the others aren’t first steps:

Azure connector: for Azure resources, not Microsoft 365 audit data.

User enrichment settings: enrich alerts with user data, not data ingestion.

Automatic log upload settings: not the initial integration step for Cloud Apps data.

Nairobi, Kenya

Legofi Commented on April 15, 2026
It's helpful
Thabazimbi, South Africa

Community-Verified AI Explanation Commented on April 15, 2026
Question 20:
Correct answer: D. Routing table

Why: The routing table is volatile data stored in memory. Capturing it first preserves the current network paths and possible attacker movement before isolating the server. Powering off or seizing the host could cause this information to be lost.

Why not the others:

- A Hard disk and B Primary boot partition: non-volatile; should be collected after preserving volatile data. - C Malicious files: important, but not as time-sensitive as live routing info. - E Static IP address: configuration data; less critical than the live routing state for immediate incident reconstruction.

Key concept: In incident response, collect volatile data first (memory, routing table, active connections) to preserve evidence before performing actions that could alter or destroy it.

Half Way Tree, Jamaica

Community-Verified AI Explanation Commented on April 15, 2026
Question 19:
Correct answer: C. Reverse engineering
Why: A malicious binary lacks source code, so the best way to understand its behavior is to perform reverse engineering—disassembling, decompiling, and tracing its code paths to reveal payloads, persistence, and IOCs.
Why not the others:

Code analysis: requires source code; not applicable to a compiled binary.

Static analysis: analyzes code without execution, but for binaries this is limited and often less revealing than reverse engineering.

Fuzzing: tests how software handles malformed inputs; not designed to reveal the internal logic or capabilities of a malware binary.

Key concept: For binaries, reverse engineering is the primary method to uncover how malware works and what indicators to hunt for.
Half Way Tree, Jamaica

Community-Verified AI Explanation Commented on April 15, 2026
Following a recent power outage, a server in the datacenter has been constantly going offline and losing its configuration. Users have been experiencing access issues while using the application on the server. The server technician notices the date and time are incorrect when the server is online. All other servers are working.

Correct answers: A and B.

Why:

- A. The server has a faulty power supply — Inconsistent power can cause abrupt shutdowns/restarts and loss of configuration data. - B. The server has a CMOS battery failure — The CMOS/RTC stores BIOS settings and the real-time clock. If the battery is failing, power loss can reset BIOS settings and the system clock, causing configuration loss and incorrect date/time after outages.

Other options (OS updates, LED panel, NTP) are less likely to explain both ongoing outages and time/date reset after a power event. Once you replace the CMOS battery and ensure stable power (with a capable PSU/UPS), reconfigure BIOS and enable proper time synchronization.

Johannesburg, South Africa

Salesforce Certified AI Associate Exam Questions Salesforce Certified AI Associate

Pass the Salesforce Certified AI Associate exam with community-verified questions and a free, built-in AI Tutor.

SalesforceSalesforce-AI-AssociateSalesforce Certified AI Associate Exam

Total Questions: 102

Certified AI Associate Exam Discussions & Posts

Salesforce Certified AI Associate Exam Questions
Salesforce Certified AI Associate

Salesforce
Salesforce-AI-Associate
Salesforce Certified AI Associate Exam