ATLASSIAN ACP-520 Exam Prep
Atlassian Cloud Organization Admin Certification (Page 2 )

• B — Incorporate contextual output encoding/escaping: Encode untrusted data based on its context—HTML, attribute, JavaScript, or URL—so the browser treats it as data rather than executable code.
• D — Run untrusted HTML input through an HTML sanitization engine: Sanitization removes or neutralizes dangerous elements and attributes, such as script-related content, while preserving safe HTML where needed.

• A: Enabling client-side scripts does not prevent XSS. Script restrictions may reduce risk in some environments, but “enable” is not a general XSS control.
• C: Cookie inspection is unrelated to preventing injected scripts.
• E: The SameSite cookie attribute primarily helps mitigate CSRF, not XSS. It should not generally be described as an XSS prevention measure.

• User-generated content: precisa ser coletado, filtrado, anonimizado e rotulado.
• Moderation logs: podem conter dados inconsistentes, informações pessoais e decisões que precisam ser revisadas.
• Content moderation guidelines: são regras ou orientações, não dados diretamente utilizáveis para medir o comportamento do modelo.

• Which batches and markets are affected
• Whether the deficiency affects product quality, efficacy, or patient safety
• Whether the issue is an ongoing manufacturing problem
• Whether immediate containment or distribution suspension is needed

• A — Withdraw the product: may be necessary, but is premature before assessing the risk.
• B — “Dear Dr.” letter: may be appropriate later, depending on the risk.
• C — Notify global authorities: may become mandatory, but the scope and urgency depend on the assessment.

• A — VARIANT "null" values are not the same as SQL NULL values: True

• B — Frequent, single-row DMLs: False

• C — Validate data before loading into the target table: True

• D — Use staging tables to manage MERGE statements: True

• Additional running costs: $800
• Less depreciation: $300

• The $100 reduction in fixed production overhead absorbed is only an accounting reallocation. It does not represent an actual saving in total fixed overhead.

• A. Continue using the current system: It will eventually fail and offers no recovery strategy.
• B. Virtualize the entire system: Correct. A VM can preserve the old OS, application, settings, and dependencies while running on newer physical hardware.
• C. Containerize the application: Containers generally share the host OS kernel and are not suitable for preserving an entire incompatible legacy OS.
• D. Clone the disks to USB: A disk clone may provide backup or temporary recovery, but it does not solve hardware compatibility or provide a reliable long-term platform.

• Packet capture: Collects and analyzes network traffic; it does not describe the scan type.
• Privilege escalation: Attempts to gain higher permissions after obtaining access.
• System enumeration: Identifies systems, services, or resources, but is not specifically defined by the lack of credentials.
• Passive scan: Observes traffic without actively probing systems. A vulnerability scan is generally an active assessment.

• Due diligence: Researching and evaluating a third party before entering or continuing a business relationship.
• Internal audit: Evaluates controls within the organization itself.
• Penetration testing: Actively tests systems for exploitable vulnerabilities.
• Attestation: A formal statement or report—such as a SOC 2 report—provided by an independent party about an organization’s controls.

• Warehouse size: Larger warehouses use more credits per unit of time than smaller ones.
• Number of clusters: A multi-cluster warehouse can run multiple clusters, and each active cluster consumes credits. Therefore, increasing the maximum or active cluster count can increase consumption.

• Number of users: More users may cause more workload, but users themselves do not consume credits.
• Amount of data processed: Snowflake warehouses are not charged directly by the amount of data scanned. However, processing more data may cause the warehouse to run longer, indirectly increasing credits.

• Third-party enterprise application with embedded AI: The provider manages most security controls. The customer mainly manages users, data, and configuration.
• Application using an existing third-party FM: The customer controls the application and its inputs/outputs, but the provider controls the underlying model and infrastructure.
• Fine-tuning a third-party FM: The customer has additional responsibility for training data, data protection, and model behavior, but still depends on the provider’s base model and platform.
• Building and training a model from scratch: The customer controls the model, training data, application, deployment, and many security safeguards. This creates the greatest security ownership.

• The selling division receives a credit based on a favourable transfer price, often market price or full cost plus a profit margin.
• The receiving division is charged a lower amount, such as marginal cost or another agreed price.
• The difference is accounted for centrally by head office or a corporate adjustment.

• 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.

• Firewalls and routers
• Servers and applications
• Endpoint devices
• Authentication systems
• Cloud services

• Centralized log collection
• Event correlation
• Dashboards and reporting
• Alerting
• Compliance support

• Information sources: Collect security-related data, such as network packets, system logs, audit records, or application events.
• Analysis: Examines the collected data to identify suspicious activity, attacks, or policy violations.
• Response: Takes or recommends action, such as generating alerts, logging the event, or supporting a protective response.

• Third-party attestation: Provides independent validation, but is typically periodic and performed by an external party.
• Penetration testing: Simulates attacks to identify exploitable vulnerabilities; it does not primarily track compliance.
• Vulnerability scans: Identify technical weaknesses, but do not provide a detailed review of procedures or compliance activities.

• B. Versioned configuration files — Infrastructure is defined in machine-readable files and tracked in version control, allowing repeatable provisioning and change history.
• D. Docker files — Assuming “Doctor files” is a typo for Dockerfiles, a Dockerfile declaratively defines how to build a container image and is commonly considered infrastructure/configuration as code.

• A. Cloned virtual machine images — These are prebuilt infrastructure artifacts, but cloning an image is not itself a code-based infrastructure definition.
• C. Change management database records — A CMDB records information about infrastructure; it does not define or provision that infrastructure.

• Employees can work from any location, including remote sites and offices worldwide.
• Login can be relatively seamless after the device is registered.
• It does not require distributing additional physical authentication equipment.

• Geotagging: Identifies location, but location alone is not a strong authentication method and may prevent legitimate remote access.
• Smart cards: Provide strong authentication but require card readers and physical cards, increasing equipment and management requirements.
• Time-based logins: Typically involve one-time codes that users must enter, making the experience less seamless and potentially requiring a token or authenticator app.

• Labels control access and protection, not blanket exclusion.
• If the user has permission to open a labeled file—especially if the label allows authenticated users to view it—Copilot can use that file in a summary.
• If the label applies encryption and the user lacks the required usage rights, Copilot should not use the file or reveal its contents.
• Copilot honors the user’s existing permissions, so it should not expose labeled content to an unauthorized user.

• DB1 (500 TB): Azure Data Box Heavy

• Virtual machines: Azure Migrate

• Lead investigator: Coordinates the investigation and directs the response.
• Information security representative: Provides security expertise and helps assess the incident.
• Technical representative: Preserves system evidence, takes forensic backups, and supports technical analysis.
• Legal representative: Advises on legal, regulatory, privacy, and evidence-handling requirements.

• STIX: A standardized format for exchanging cyber-threat intelligence.
• OVAL: A language and framework for describing and assessing system vulnerabilities and configuration issues.
• Threat intelligence feed: Provides current indicators or threat information, but is not specifically a reference for common application exploitation techniques.
• CVE: A catalog of publicly disclosed vulnerabilities, identified by unique CVE numbers.

• One physical NIC is active and carries traffic.
• Another NIC remains on standby.
• If the active uplink fails, traffic switches to the backup uplink.
• This provides redundancy without requiring special switch-side link aggregation configuration.

• Balance-SLB: Distributes traffic based on source load balancing; not the default.
• Balance-TCP: Uses dynamic TCP traffic balancing and generally requires appropriate network-switch configuration.
• No Uplink Bond: Incorrect because vs0 is normally configured with redundant uplinks.

• One network address
• Two usable host addresses—one for each router
• One broadcast address

• Router A: 10.2.1.1
• Router B: 10.2.1.2

• Azure Machine Learning Studio is designed to build, test, train, and deploy machine-learning models, including predictive analytics.
• Azure Cosmos DB is a globally distributed NoSQL database. It can store data used by an AI solution, but it is not the primary service for developing and deploying predictive models.
• Therefore, the appropriate recommendation is Azure Machine Learning Studio, as shown in Question 15.

• Case Assignment Rules normalmente são avaliadas quando o caso é criado. Elas podem ser executadas novamente em uma edição somente se o usuário marcar “Assign using active assignment rules”; isso não acontece automaticamente apenas porque o caso foi fechado.
• Portanto, criar uma regra de atribuição não corrige os centenas de casos já fechados e não garante, por si só, que um caso será transferido antes de ser fechado.

• B — Usar uma ferramenta de dados: corrige os registros históricos já existentes, atualizando o Owner.
• D — Criar uma regra de validação: impede que um caso seja fechado enquanto ainda pertence a uma fila. Exemplo conceitual: se Status = Closed e o proprietário for uma fila, bloquear o salvamento.

Statement 1 — Correct

Statement 2 — Correct

• The audience and stakeholders
• The complexity and type of requirements
• The communication method and level of detail needed
• Whether to use diagrams, models, documents, demonstrations, or slides

• B: Authority to host the presentation is not the key preparation activity.
• C: Stakeholder sign-off may occur later; presenting requirements does not always require formal approval beforehand.
• D: A scribe may be useful in a meeting, but hiring one is not a core preparation requirement for presenting.

• Disable high concurrency: This changes session-sharing behavior but does not provide dedicated compute or true workload separation.
• Modify the number of executors: Adjusts resources within a Spark job; it does not isolate bronze and silver processes.
• Set the default environment: Controls libraries and runtime configuration, not compute isolation.

• Confirming the data has the required columns and data types
• Handling missing or duplicate records
• Converting data into a suitable format
• Selecting or engineering useful features
• Splitting it into training, validation, and test sets

1. Prepare the training data
2. Select an algorithm or configure AutoML
3. Train the model
4. Evaluate its performance
5. Tune hyperparameters if needed
6. Deploy the model

• The job keeps 31 days of regular restore points. A backup from the week of January 1 is older than 31 days by March 20, so it is no longer expected to remain on the performance tier.
• It is also a weekly GFS full backup, retained for 14 weeks. Therefore, it is still protected and has not been deleted.
• In a Scale-out Backup Repository, older backups can be moved from the performance tier to the capacity tier—typically object storage—according to the configured offload policy.
• The archive tier is not selected because the question’s configuration does not indicate that the backup has been archived there.

• B. Conditional access policies

• D. Additional authentication factors

• Disciplinary actions: Punish users but do not prevent credential-based attacks.
• More regular account audits: Useful for finding inactive or excessive accounts, but less direct for stopping unauthorized logins.
• Content filtering: Primarily controls web or email content, not authentication.
• Review of permissions: Reduces privilege but may not stop unauthorized account access.

• Allow only approved MAC addresses
• Limit the number of learned addresses
• Shut down or restrict a port when a violation occurs

• Load balancer: Distributes application traffic; it does not protect switch MAC tables.
• IPS: May detect some attack traffic, but it is not the direct control for limiting MAC addresses on switch ports.
• NGFW: Filters routed network traffic, while this attack targets Layer 2 switch behavior.

• Reduce the training data: Usually makes overfitting worse.
• Add hyperparameters: Hyperparameters already control training; simply adding them does not solve the problem. Proper tuning could help, but that is not what the option states.
• Increase training time: More training can worsen overfitting if the model is already memorizing the training data.

• Crowd-sourced evaluation: Requires recruiting and managing many evaluators.
• Human-worker evaluation: Provides human judgment but requires coordination, labeling instructions, and quality control.
• RLHF: Is a training and alignment process, not simply a low-overhead model comparison method.
• Automatic model evaluation: Quickly evaluates outputs using predefined toxicity or quality metrics with minimal manual effort.

• Fine-tuning requires a properly formatted training dataset.
• For classification or instruction-style fine-tuning, examples are supplied in JSONL format.
• Each line represents one training example, typically containing the input prompt and the desired completion or label.
• The file must then be uploaded or made available to the Microsoft Foundry fine-tuning workflow.

• A: Prompt Flow is useful for prompt experimentation, but it is not required to start fine-tuning.
• B: You do not need to deploy the base model to an online endpoint before fine-tuning.
• C: Tracing supports observability and evaluation, but it does not prepare the fine-tuning job or dataset.

• A PostToolUse hook for file writes under src/api/ that runs the linter.
• A separate PostToolUse hook for API/data processing that normalizes payloads to snake_case.

• Virtual introductions: Team members share their role, experience, location, and working preferences.
• Online icebreakers: Short, informal activities such as “two truths and a lie.”
• Virtual coffee chats: Pair team members for informal, non-project conversations.
• Collaborative problem-solving: The team works together on an online puzzle or scenario.
• Retrospectives: The team discusses what went well, what did not, and how to improve.
• Working-agreement sessions: The team agrees on response times, meeting etiquette, communication channels, and handoffs.
• Recognition sessions: Members acknowledge one another’s contributions.
• Remote workshops: Teams jointly create plans, diagrams, or schedules using collaboration tools.

• Containment policy: controls permitted communication during containment.
• Detection/response configuration: determines when containment is initiated.

• Plenum: Describes a fire-resistant cable jacket rating, not a transmission medium or speed.
• Ethernet: Describes a networking technology; Ethernet can use copper or fiber, so the term alone does not identify the fastest cable type.
• DAC (Direct Attach Copper): Uses copper and is typically intended for short-distance, high-speed connections in data centers.

• A: Concerns task scheduling and avoiding missed runs, not directly sizing the warehouse.
• B: The amount of data in a stream alone does not reliably determine compute needs; the SQL processing is what matters.
• D: A multi-cluster warehouse addresses concurrency, not the compute size needed for one task execution.

• Dark web: A place where leaked credentials may be sold or shared, not usually the original source.
• Threat feeds: Collections of intelligence about threats, not a typical leakage source.
• State actors: A type of attacker, not an accidental corporate disclosure source.
• Vulnerability databases: Track known vulnerabilities, rather than leaked credentials.

• A — Tables
• B — Named file formats
• C — Schemas
• E — Databases

Why these are correct

• Tables: Can be cloned using CREATE TABLE ... CLONE. The clone initially uses zero-copy cloning, so storage is shared until changes are made.
• Named file formats: Can be cloned as database objects, preserving the file-format definition.
• Schemas: A schema and its eligible objects can be cloned.
• Databases: A database and its eligible schemas and objects can be cloned.

Why the others are incorrect

• Shares: A share is an account-level data-sharing object, not a normal cloneable database object.
• Users: Users are security principals and cannot be cloned with CLONE.

• Report on the discount approval time for quotes

• Dashboard of opportunity stage duration

• Lead conversion trends measure lead-generation effectiveness, not opportunity closing time.
• Campaign ROI evaluates marketing performance, not delays in the sales process.

• Business risks
• Potential financial, operational, legal, or reputational impact
• How governance reduces risk to organizational objectives
• How much risk the organization is willing to accept

• 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.

• Actions that call external APIs
• Subflows
• Connection and credential aliases
• Supporting integration logic

• Action — A reusable operation within a flow, such as “Create a record” or “Send email.”
• Connection — Defines how ServiceNow authenticates and communicates with an external system.
• Integration step — A general flow step, not the name for the packaged integration.

• Type 1 hypervisor: Runs directly on the physical hardware (“bare metal”).
• Type 2 hypervisor: Runs inside the host OS as an application.

• Bare-metal installation: Type 1
• Local management only: Not a defining characteristic
• Specific hardware requirements: Not the key distinction

• Active remediation actions role in Microsoft Defender for Endpoint

• Security Reader role in Azure AD/Entra ID

• Security Administrator is broader than necessary because it grants extensive security-management permissions, violating least privilege.
• Compliance Data Administrator is intended for compliance data management, not Defender for Endpoint remediation approvals.

• Determine risk tolerance: Tolerance defines how much variation or exposure the organization is willing to accept. Scenarios may help apply tolerance, but they do not primarily determine it.
• Determine risk appetite: Risk appetite is established at an enterprise or governance level. Scenarios help compare risks with the appetite; they do not normally establish it.
• Develop risk responses: Once risks are identified and assessed, scenarios support response planning. However, response development comes later.