Salesforce Certified AI Specialist Exam Prep
Certified AI Specialist (Page 4 )

• A: It allows for branching and merging...

• C: It allows multiple engineers to work against the same code and configuration files and manage differences and conflicts.

• B: Build automation and infrastructure provisioning are handled by CI/CD and automation tools, not by version control alone.
• D: User Stories and backlogs are managed by project-management tools.
• E: Version control stores and tracks code, but it does not itself ensure that developers write effective unit tests.

• Tokens: Individual units of text processed by a model, such as words, subwords, or characters.
• Models: Systems that learn patterns and generate predictions or outputs.
• Binaries: Data represented using the binary number system, not semantic representations.

• Customer: Dual
• Date: Dual
• Sales: DirectQuery
• SalesAggregate: Import

• Sales – DirectQuery: Near-real-time data should not rely on an imported cache. DirectQuery sends queries to the source when visuals are loaded, allowing recent sales data to appear.
• SalesAggregate – Import: A weekly refresh is sufficient, and imported data generally gives faster visual performance because it is stored in Power BI’s in-memory cache.
• Customer and Date – Dual: These are dimension tables used with both the DirectQuery Sales table and the imported SalesAggregate table. Dual allows Power BI to use cached data when possible, while still behaving appropriately in queries involving DirectQuery data. This can improve performance and reduce unnecessary limited relationships.

• The message came from the claimed sender (authentication)
• The message was not altered in transit (integrity)
• The sender cannot easily deny having signed and sent it (non-repudiation)

• Compliance standards: May be a reason in some organizations, but it is not the primary security benefit.
• Increase delivery rates: Digital signatures do not generally improve email delivery.
• Block phishing attacks: Signatures can help recipients verify legitimate messages, but they do not block phishing by themselves.

• T?o m?t Systems Manager Automation runbook d? ch?y l?nh kh?i d?ng l?i ?ng d?ng trên các EC2 instance.
• T?o m?t Amazon EventBridge rule ph?n ?ng khi CloudWatch alarm chuy?n sang tr?ng thái ALARM.
• Ð?t Automation runbook làm target c?a EventBridge rule.

1. Log c?a ?ng d?ng du?c ghi vào CloudWatch Logs.
2. CloudWatch alarm phát hi?n nhi?u l?i liên quan d?n vi?c restart.
3. Alarm chuy?n sang tr?ng thái ALARM.
4. EventBridge nh?n s? ki?n thay d?i tr?ng thái này.
5. EventBridge g?i Systems Manager Automation runbook.
6. Runbook restart ?ng d?ng trên các EC2 instance, không restart c? instance.

• A: SNS không tr?c ti?p g?i Systems Manager Automation runbook nhu m?t target native.
• B: Lambda có th? th?c hi?n vi?c restart, nhung ph?i t? x? lý vi?c xác d?nh instance và ch?y l?nh t? xa.
• C: Lambda ch? làm trung gian d? g?i runbook, không c?n thi?t vì EventBridge có th? g?i runbook tr?c ti?p.

• Due diligence: Investigating and evaluating a third party before entering into or continuing a business relationship.
• Attestation: An independent party provides assurance about an organization’s claims or controls. A SOC 2 examination is an example of an attestation engagement, but the analyst requesting and reviewing the report is performing due diligence.
• Internal audit: Reviews the organization’s own internal controls.
• Penetration testing: Actively tests systems for exploitable vulnerabilities.

apex 
for (Account updatedAccount : Trigger.new) { 
    AccountTriggerHelper.notinyxternalSystem(updatedAccount.Id); 
}

• Too many callouts: The callout occurs asynchronously in the future method, and each future transaction can perform its permitted callouts.
• Too many SOQL queries: The SOQL query is inside the future execution, not repeatedly in the trigger transaction.
• Callout from triggers not supported: Direct callouts from triggers are prohibited, but the trigger delegates the callout to asynchronous Apex, which is the correct pattern.

• Scheduling overlapping working hours for critical collaboration
• Assigning independent tasks that can proceed asynchronously
• Adding clear handoffs between offices
• Documenting decisions, questions, and required inputs
• Building appropriate time into the schedule for responses

• Azure Resource Manager (ARM) templates
• Or, in newer Azure terminology, Bicep files

• Asynchronous inference is designed for applications that submit individual inference requests, place them in a queue, and retrieve results later. It is useful when requests may take a long time or when traffic is intermittent.
• Batch transform is designed specifically to run inference over an entire large dataset stored in Amazon S3 and produce an output dataset.

• Large dataset/offline processing ? Batch transform
• Large or slow individual requests processed later ? Asynchronous inference

• Personally identifiable information (PII) detection

• Prompt Shields

• Self-harm content filtering addresses self-harm-related content.
• Violence content filtering addresses violent content.
• Neither protects customer PII or detects document-based prompt injection.

• Prevention costs – costs incurred to stop defects occurring, such as staff training, quality planning and supplier evaluation.
• Appraisal costs – costs of checking whether products meet standards, such as inspection, testing and quality audits.
• Internal failure costs – costs of defects found before delivery to customers, such as scrap, rework and retesting.
• External failure costs – costs of defects found after delivery, such as warranty claims, returns, complaints and product recalls.

• Training or coaching
• Mentoring
• Additional support or resources
• Targeted performance improvement plans

• A: Recognition may be appropriate, but awards do not address identified gaps.
• C: Rotating schedules does not necessarily improve the specific competencies that are missing.
• D: Assessment results should be handled sensitively, but keeping them entirely private prevents useful feedback and development discussions.

• The users currently have Office 365 Enterprise E3 licenses assigned directly.
• Assigning Microsoft 365 Enterprise E5 through a group does not automatically remove their directly assigned E3 licenses.
• From the Licenses blade, you can select the E3 product and remove it from all assigned users in one operation. This requires less effort than editing 2,500 users individually.

• Identity Governance manages access reviews, entitlement management, and related governance—not bulk license removal.
• Set-AzureAdUser does not remove product licenses.
• Set-WindowsProductKey applies Windows product keys and is unrelated to Microsoft 365 licensing.

• Existing expected ROI: 15%
• New project ROI: 14%
• Cost of capital: 12%

• Project return = 14%
• Cost of capital = 12%
• It earns 2% more than the cost of capital, so it should increase shareholder value.

• The S3 bucket as an origin for static content
• The Application Load Balancer (ALB) as an origin for dynamic content
• An Amazon Route 53 alias record pointing the application domain to CloudFront

Why this works

• Static data: CloudFront caches objects from S3 at edge locations, reducing latency for users worldwide.
• Dynamic data: CloudFront forwards requests that require application processing to the ALB. It can also maintain persistent connections and use the AWS global edge network to improve delivery performance.
• One domain: CloudFront behaviors can route different path patterns to different origins—for example, /static/* to S3 and other paths to the ALB.

Why the other options are incorrect

• Global Accelerator is designed primarily for accelerating traffic to applications with endpoints such as ALBs, not for caching S3 content.
• Using separate domains or splitting static and dynamic traffic adds unnecessary complexity.
• CloudFront already provides the required global edge distribution and caching, so combining it with Global Accelerator is unnecessary here.

1. Define – establish the project’s objectives, scope, feasibility and requirements.
2. Design – plan how the project will be carried out, including resources, schedules and controls.
3. Deliver – execute the project and produce the required outputs.
4. Develop – close the project formally and capture learning for future projects.

• Filing documentation and records
• Closing project and accounting systems
• Settling costs
• Handing over the completed work and obtaining sign-off
• Reviewing performance and giving feedback

• Discovers sensitive data in Amazon S3 buckets
• Uses machine learning and pattern matching to identify data such as personally identifiable information (PII)
• Helps assess S3 security and privacy risks
• Generates findings for potentially sensitive or exposed data

• Amazon Detective: Investigates security findings and performs root-cause analysis.
• Amazon GuardDuty: Detects threats and suspicious activity across AWS accounts and workloads.
• AWS IAM Access Analyzer: Identifies resources that are accessible from outside their intended trust zone, such as publicly accessible S3 buckets.

• Sensitive document shared during a Teams chat: Teams chat
• Sensitive document stored/shared in a Teams channel: Teams channel messages

• Files shared in a chat are generally stored in the sender’s OneDrive, but the relevant DLP context is the Teams chat message and its attachment.
• Files shared in a channel are stored in the team’s SharePoint site, but the relevant context is the Teams channel message.
• The DLP rule can then use an action such as restrict access to the content to prevent guest users from opening the sensitive document.

• Unfamiliar locations or IP addresses
• Anonymous or malicious IP addresses
• Leaked credentials
• Unusual sign-in behavior

• Reviewing security policies and standards checks whether controls are documented, not whether they work.
• Reviewing security logs may show that monitoring is occurring, but does not directly establish that preventive controls are effective.
• Analyzing technical security practices is broad and may assess implementation, but it is less objective than an actual controlled attack.

• Design/compliance effectiveness: policies, standards, and procedures
• Operational effectiveness: logs, records, and evidence that controls operate
• Technical security effectiveness: penetration testing or vulnerability testing

• C series: CPU-optimized instances; not designed primarily for LLM training.
• G series: GPU instances generally optimized for graphics and some machine learning workloads, but not specifically large-scale model training efficiency.
• P series: Powerful GPU instances for ML training, but typically consume more energy than Trainium-based instances for suitable training workloads.

• SQL injection
• Cross-site scripting (XSS)
• Malicious or malformed web requests
• Some application-layer attacks

• Layer 4 firewall: Filters based mainly on IP addresses, ports, and transport protocols; it cannot deeply inspect web application content.
• NGFW: Provides advanced network filtering and may include application awareness, but a WAF is more specifically designed to protect web applications.
• UTM: Combines several security functions in one appliance, but it is not the most targeted solution for a web portal.

• Install unauthorized or unsupported software
• Access features blocked by the manufacturer or operating system
• Customize system behavior beyond normal settings

• Digitally signing the software: Verifies authenticity and integrity. It does not hide how the software works.
• Limiting third-party libraries: May reduce dependencies and attack surface, but does not directly prevent reverse engineering.
• Using compile flags: Some compiler options can remove debugging information or optimize code, which may provide limited protection, but this is not the primary technique identified for preventing reverse engineering.

• Accept: Acknowledging the risk without taking additional action.
• Transfer: Shifting the risk’s impact to a third party, such as through insurance or outsourcing.
• Mitigate: Reducing the likelihood or impact of the risk through security controls.
• Avoid: Eliminating the activity or process that creates the risk.

• An Adj-SID does not necessarily need manual configuration.
• It can often be dynamically allocated by the router, although manual configuration may also be possible.
• The adjacency information is advertised through the IGP, such as IS-IS or OSPF.

• It is identified by an Adj-SID.
• It represents a particular adjacency of a node.
• Its meaning is generally local to the advertising node; unlike a Node-SID, it is not a globally reusable instruction across the network.

• Azure Cosmos DB is a globally distributed NoSQL database service. It stores and retrieves application data, but it does not provide the complete environment for building, testing, and deploying predictive analytics models.
• Azure Machine Learning is the appropriate service for this requirement. It supports creating experiments, training and testing models, and deploying them for predictions.
• Azure Machine Learning Studio provides the browser-based interface for these tasks.

sql 
ALTER TABLE dbo.DimCustomer 
ADD CONSTRAINT PK_DimCustomer 
PRIMARY KEY NONCLUSTERED (CustomerKey) 
NOT ENFORCED;

1. ALTER TABLE dbo.DimCustomer
2. ADD CONSTRAINT PK_DimCustomer PRIMARY KEY NONCLUSTERED (CustomerKey)
3. NOT ENFORCED

• Wireshark captures packets and allows you to examine their contents, potentially revealing usernames and passwords sent without encryption.
• Netcat creates network connections and can send or receive data, but it is not primarily a packet-analysis tool.
• Nessus scans for known vulnerabilities and configuration issues; it generally would not prove that credentials are visible in transit.
• Nmap performs network discovery and port/service scanning, not detailed packet-content inspection.

• CCTV system — Detective control: Cameras monitor and record activity, helping identify or investigate security incidents after—or while—they occur.
• Signs about being filmed — Deterrent control: The signs discourage people from attempting unauthorized or criminal activity because they know surveillance is present.

• Preventive: Stops an incident from occurring, such as a lock or access-control system.
• Corrective: Restores systems after an incident.
• Directive: Instructs people to behave in a certain way, such as a security policy.
• Compensating: Provides an alternative control when the preferred control cannot be implemented.

• Accept: Acknowledge the risk and take no additional action.
• Transfer: Shift the risk or financial impact to another party, such as through insurance or a contract.
• Mitigate: Reduce the likelihood or impact using controls, such as firewalls, training, or backups.
• Avoid: Eliminate the activity that creates the risk.

• Tracking who accessed or changed customer information
• Detecting unauthorized activity
• Supporting accountability and investigations
• Providing evidence for compliance and incident response

• Single sign-on not enabled: SSO improves convenience and centralized authentication, but it is not essential for access control.
• Complex passwords not required: This weakens authentication, but other controls may reduce the risk.
• Security baseline not consistently applied: A concern, but it is less specific and less immediately damaging than having no activity records at all.

• Prevent connections to malware command-and-control servers
• Identify infected systems making those DNS requests
• Collect information for incident response

• A: Attackers do not normally use sinkholes to map a company’s domain structure.
• B: Redirecting employees from legitimate sites to attacker-owned sites describes malicious DNS manipulation, not a defensive sinkhole.
• D: Attracting attackers away from real resources is more closely associated with a honeypot or deception system.

• Restrict a port to a defined number of MAC addresses
• Use statically assigned or “sticky” MAC addresses
• Shut down or restrict the port when violations occur

• Load balancer: Distributes application traffic; it does not protect switch MAC tables.
• IPS: May detect suspicious network activity, but is not the primary control for limiting switch MAC entries.
• NGFW: Filters traffic at the network boundary; it does not directly prevent MAC table flooding on access switches.

• Honeypots: Deliberately vulnerable or attractive systems designed to lure attackers. Legitimate users should not access them, so activity on a honeypot is a strong indicator of compromise.
• DNS sinkholes: Redirect requests for known malicious domains to a controlled internal system instead of allowing the connection. This can identify infected hosts attempting to contact command-and-control infrastructure.

• Tokenization: Replaces sensitive data with non-sensitive tokens; it protects data but does not primarily detect attackers.
• CI/CD: Automates software development and deployment processes.
• Threat modeling: Identifies potential threats and weaknesses before or during system design.
• Data obfuscation: Makes data harder to understand, but it is not an attacker-detection method.

• A. IDS attack signatures: An IDS primarily detects suspicious network activity; it is not the best control for scanning uploaded documents for malware.
• C. Load balancer: Distributes traffic among servers and improves availability, but does not inspect files for malware.
• D. WAF: Protects web applications against attacks such as SQL injection and command injection. It may help secure the upload function, but malware scanning is the key requirement here.

• Load balancer: Distributes application traffic; it does not protect switch MAC tables.
• IPS: May detect some attack traffic, but it is not the primary control for switch-port MAC learning.
• NGFW: Protects at network boundaries but does not directly limit MAC addresses learned on access ports.

• Menerjemahkan soal ke bahasa Indonesia.
• Menyebutkan jawaban yang benar.
• Menjelaskan alasan jawaban tersebut.
• Memberi tahu jika kunci jawaban yang tersedia tampaknya keliru.
• Menjelaskan istilah teknis penting dalam soal.

1. Define Power Platform DLP policies

1. Add custom prompt modifications

1. Conduct adversarial testing

• The product is already in production.
• The organization is examining the product’s functions and cost drivers.
• The objective is to reduce cost without reducing perceived value or required quality.

• Kaizen costing focuses on continuous, usually incremental, cost reductions during the manufacturing stage. It is not specifically the systematic examination of a product’s functions.
• Standard costing compares actual costs with predetermined standards; it is a control technique, not a product redesign method.
• Process innovation concerns developing significantly improved production processes, rather than analysing the product’s functions and value.

• Understating expenses and liabilities
• Overstating profit
• Misleading users of the financial information
• Potentially breaching integrity, objectivity, and professional competence and due care

• B. Deploy to a SageMaker endpoint: This is needed for SageMaker-hosted models, not for a model intended to be used directly in Amazon Bedrock.
• C. Register with SageMaker Model Registry: SageMaker Model Registry is unrelated to invoking custom models in Bedrock.
• D. Grant access in Bedrock: Access permissions may be necessary, but they do not replace the required provisioned capacity.

• Amazon SageMaker — charges apply for compute, storage, and other resources used.
• AWS Config — generally charges for configuration recording and rule evaluations.
• Amazon CloudWatch — has a limited free tier, but many metrics, logs, alarms, and API usage can incur charges.

• Reporting relationships
• Organizational structures
• Leadership styles
• Team maturity
• Individual and group behavior
• Communication and motivation

• Acquire Project Team: Focuses on obtaining the people needed for the project.
• Develop Project Team: Focuses on improving team skills, relationships, and performance.
• Manage Project Team: Focuses on tracking performance, resolving issues, and managing changes within the team.

• The vulnerability scan identifies a vulnerability.
• The administrator verifies that the vulnerability is not present.
• Therefore, the scan produced a false positive.

• True positive: A vulnerability is reported and genuinely exists.
• False negative: A vulnerability exists, but the scan fails to detect it.
• True negative: No vulnerability exists, and the scan correctly reports no issue.

• A — Too infrequent: Formal strategies are often reviewed only every three to five years. This can make the organisation slow to respond to rapid changes in technology, competitors, regulation or customer needs.

• C — Relies on predicting the future: The approach assumes senior managers can make reasonably reliable forecasts and assumptions. In uncertain environments, these assumptions may quickly become invalid.

• E — Can restrict radical innovation: Formal planning tends to favour existing information, established objectives and incremental improvements. This may discourage experimental or highly innovative strategies that emerge from lower levels of the organisation.

• B concerns the difficulty of achieving competitive advantage through core competences when competing with focused specialists. It is not specifically a drawback of the formal top-down planning process.
• D is an activity organisations should undertake—continuous monitoring—not a drawback of the approach itself.

• Identity provider (IdP): Authenticates users and asserts their identity.
• Service provider (SP): The application being accessed—in this case, Salesforce/the employee portal.
• Identity store: A directory that stores user identities, such as Active Directory.
• Authentication store: A system used to validate credentials.

• Accept: Acknowledging the risk and choosing not to take additional action.
• Transfer: Shifting the risk or financial consequences to another party, such as an insurer.
• Mitigate: Reducing the likelihood or impact through controls, such as patching or access restrictions.
• Avoid: Eliminating the activity that creates the risk.

• Documented and evaluated
• Assessed for security, cost, and operational impact
• Reviewed and approved by the appropriate personnel
• Tested before implementation
• Tracked with version history and rollback procedures

• Risk management process: May help assess the issues, but it does not authorize and manage design changes.
• Product design process: Applies primarily while creating a new design.
• Design review process: Normally evaluates a design before release; this design is already released.