Governance, Security, Ethics, and Future Directions

Introduction and Context

Supply chain management software has evolved from record-keeping applications into interconnected platforms that coordinate planning, procurement, logistics, manufacturing, compliance, and customer service. Enterprise resource planning (ERP), warehouse management systems (WMS), transportation management systems (TMS), supplier-management platforms, analytics tools, application programming interfaces (APIs), Internet of Things (IoT) devices, and artificial-intelligence (AI) services now exchange data across organizational boundaries.

This integration creates value, but it also creates exposure. A compromised supplier account can disrupt production; inaccurate master data can generate inappropriate purchases; an opaque algorithm can disadvantage a supplier; and a poorly governed cloud integration can disclose personal or commercially sensitive information. Governance is therefore not an administrative afterthought. It is the system by which an organization directs technology, assigns accountability, controls risk, demonstrates compliance, and preserves stakeholder trust.

This chapter examines governance, cybersecurity, privacy, resilience, auditability, model risk, responsible AI, and regulatory compliance in supply chain software. It then evaluates blockchain, industry clouds, sustainability platforms, autonomous ecosystems, and the workforce capabilities required for the next generation of supply chain management.

Learning Objectives

By the end of this chapter, the reader should be able to:

1. Design a governance structure for access, data, technology, models, and third parties.
2. Explain major cybersecurity, privacy, resilience, and auditability controls.
3. Assess model risk and apply principles of responsible AI in supply chain decisions.
4. Relate supply chain software controls to regulatory and contractual obligations.
5. Critically evaluate blockchain, industry clouds, sustainability platforms, and autonomous ecosystems.
6. Recommend workforce capabilities and implementation practices for future supply chains.

1. Governance as an Operating System for Accountability

Governance is the allocation of decision rights, responsibilities, controls, and escalation mechanisms. It differs from management. Management operates processes and delivers results; governance determines who may decide, under what rules, with what evidence, and subject to whose oversight.

A useful governance design has at least six domains:

A governance charter should specify decision rights, risk appetite, mandatory controls, exception procedures, reporting frequency, and consequences for non-compliance. It should also distinguish accountability from execution. An IT administrator may configure a role, but the business process owner remains accountable for whether that role is appropriate.

Three Lines of Responsibility

A practical assurance model uses three lines. The first line consists of operational managers who own risks and controls. The second line consists of risk, compliance, privacy, and security specialists who establish standards and challenge operational decisions. The third line is internal audit, which provides independent assurance to senior leadership or the board. External auditors and regulators provide additional, independent scrutiny but do not replace management accountability.

Governance should be risk-based rather than control-maximalist. A low-risk informational dashboard need not receive the same approval process as an algorithm that releases payments, allocates scarce inventory, or makes employment decisions. However, risk-based governance must not become a justification for undocumented exceptions.

2. Identity, Access, and Segregation of Duties

Supply chain platforms contain valuable data and powerful transactions. Identity and access management (IAM) establishes who a user or service is; authorization determines what that identity may do. Modern governance favors least privilege: users and services receive only the access necessary for an authorized task, for only as long as necessary.

Important controls include:

• single sign-on and phishing-resistant multifactor authentication;
• role-based access control, supplemented by attributes such as location, supplier relationship, or transaction value;
• privileged-access management with time-limited elevation;
• joiner, mover, and leaver procedures;
• quarterly or risk-based access recertification;
• service-account ownership, credential rotation, and secrets management;
• segregation of duties (SoD); and
• immutable logging of authentication and high-risk transactions.

SoD prevents incompatible activities from being controlled by one person or one identity. For example, the same individual should ordinarily not create a supplier, change its bank details, approve an invoice, and release payment. Automated workflows can enforce dual approval, but automation does not eliminate the need to review the design of the workflow.

A common failure is permission accumulation. A user changes roles but retains former privileges, creating excessive access. Another is the use of shared accounts, which weakens accountability and makes forensic investigation difficult. Emergency accounts may be necessary for continuity, but they should be strongly protected, logged, reviewed after use, and disabled when not required.

3. Cybersecurity in the Extended Supply Chain

Supply chain security extends beyond the organization’s own network. Suppliers, logistics providers, software vendors, cloud hosts, contractors, and connected devices may all create attack paths. Threats include ransomware, credential theft, business-email compromise, software supply-chain attacks, data exfiltration, denial of service, malicious insiders, and manipulation of operational data.

A layered control architecture includes:

1. Govern: establish risk appetite, policies, asset ownership, supplier requirements, and reporting.
2. Identify: maintain inventories of applications, interfaces, data, devices, suppliers, and critical processes.
3. Protect: apply secure configuration, encryption, patching, training, network segmentation, backups, and access controls.
4. Detect: monitor anomalous logins, unusual transactions, malware indicators, data movement, and service degradation.
5. Respond: contain incidents, communicate decisions, preserve evidence, and coordinate with suppliers and authorities.
6. Recover: restore services, validate data integrity, learn from the event, and revise controls.

This structure is consistent with the broad logic of the NIST Cybersecurity Framework 2.0, although organizations must tailor implementation to their context. ISO/IEC 27001 provides a certifiable information-security management-system approach. Neither a framework nor a certification guarantees security; effectiveness depends on implementation, testing, and adaptation.

Third-Party Risk

Supplier due diligence should consider ownership, services, data handled, geographic exposure, subcontractors, software dependencies, incident history, recovery capability, and control evidence. Contracts should address security standards, breach notification, audit rights, vulnerability disclosure, data location, subcontracting, exit assistance, and deletion or return of data. Risk assessments should continue after onboarding because supplier circumstances change.

A supplier questionnaire alone is weak assurance. More reliable evidence may include independent assurance reports, penetration-test summaries, vulnerability-management metrics, observed recovery exercises, and contractual performance data. Organizations should avoid demanding controls that are impossible for smaller suppliers to implement; proportional requirements and shared support can improve ecosystem security.

4. Privacy and Data Governance

Supply chain systems process personal data about employees, drivers, customers, supplier contacts, and sometimes consumers. They may also process sensitive commercial information such as prices, forecasts, production capacity, and sourcing strategies.

Privacy governance begins with data mapping: what is collected, from whom, for what purpose, where it is stored, who receives it, how long it is retained, and how it is deleted. Key principles found in many privacy regimes include lawfulness, purpose limitation, data minimization, accuracy, storage limitation, security, and demonstrable accountability. The exact legal obligations depend on jurisdiction, role, data type, and processing activity.

Organizations should distinguish a data controller or business deciding purposes from a processor or service provider acting on instructions, while recognizing that terminology differs across laws. Cross-border transfers may require approved mechanisms and additional safeguards. Privacy impact assessments are appropriate when processing is likely to create high risk, including extensive monitoring, profiling, sensitive data use, or large-scale automated decision-making.

Privacy-enhancing techniques include pseudonymization, aggregation, tokenization, field-level access, encryption, retention limits, and controlled data-sharing environments. Pseudonymization reduces exposure but does not necessarily make data anonymous if re-identification remains reasonably possible. Encryption protects confidentiality but does not correct inaccurate data or prevent authorized misuse.

5. Resilience, Continuity, and Recovery

Resilience is the capability to absorb disruption, continue critical operations, and recover while maintaining acceptable performance and control. It includes cybersecurity recovery, but is broader: port closures, extreme weather, labor disputes, supplier failure, utility outages, geopolitical events, and poor data quality can all disrupt supply chains.

Business impact analysis identifies critical processes, dependencies, maximum tolerable disruption, and recovery priorities. Two important measures are:

• Recovery time objective (RTO): the maximum acceptable time to restore a service.
• Recovery point objective (RPO): the maximum acceptable age of restored data, expressed as a time interval.

For example, an RTO of four hours and an RPO of fifteen minutes means the service should be restored within four hours and data loss should not exceed approximately fifteen minutes of transactions, subject to the organization’s definitions and technical feasibility.

Resilience controls include redundant infrastructure, tested backups, alternate communication channels, manual workarounds, multi-sourcing, geographically separated recovery environments, and documented crisis authority. Backups must be protected from ransomware and tested through restoration, not merely reported as successful. A recovery plan that has never been exercised is an unverified assumption.

A Simple Risk-Priority Model

Organizations sometimes prioritize controls using an expected-loss approximation:

Here, E[L] is expected loss in a defined period, P(I) is the estimated probability of incident I in that period, and C(I) is the consequence if it occurs, measured in a consistent monetary or scoring unit. The model assumes that probability and consequence estimates are meaningful and that the decision context permits a simplified expected-value comparison. It does not capture cascading effects, uncertainty, legal duties, or ethical severity adequately; therefore it should support, not replace, expert judgment and scenario analysis.

6. Auditability, Data Lineage, and Control Evidence

Auditability is the ability to reconstruct what happened, who or what caused it, which data and rules were used, and whether the action complied with policy. It is essential for financial control, regulatory inspection, incident investigation, supplier disputes, and model validation.

A strong audit trail records event time, identity, source system, transaction or decision, relevant data version, authorization, change history, and outcome. Logs should be time-synchronized, access-controlled, protected from unauthorized alteration, retained according to legal and business requirements, and monitored for suspicious activity. Excessive logging can create privacy and storage risks, so collection should be purposeful.

Data lineage connects an output to its origins and transformations. For a supply plan, lineage may include demand signals, inventory balances, lead-time assumptions, supplier constraints, overrides, optimization logic, and approval events. Lineage makes it possible to distinguish a model error from a source-data error or a human override.

A control is not demonstrated merely because a policy exists. Evidence may include access reviews, exception approvals, test results, incident records, configuration baselines, recovery exercises, and reconciliations. Auditability should be designed into workflows rather than reconstructed manually at the end of the year.

7. Model Risk and Responsible AI

Supply chain software increasingly uses forecasting, anomaly detection, optimization, computer vision, natural-language interfaces, and generative AI. Model risk is the possibility that a model is wrong, misused, misunderstood, or applied outside the conditions for which it was developed.

A model-risk lifecycle includes:

1. Purpose definition: state the decision, users, affected parties, constraints, and unacceptable uses.
2. Data assessment: evaluate relevance, quality, representativeness, provenance, leakage, and rights to use.
3. Development and validation: test performance against suitable baselines and realistic scenarios.
4. Approval: obtain documented business, technical, legal, privacy, and risk review proportional to impact.
5. Deployment controls: limit access, record versions, establish human review, and define fallbacks.
6. Monitoring: track accuracy, drift, stability, fairness, security, overrides, and unintended outcomes.
7. Retirement: remove obsolete models and preserve required records.

Performance must be evaluated against the decision’s purpose. A forecast metric such as mean absolute error is not automatically appropriate for every product or cost structure. Similarly, a high classification accuracy can conceal poor performance on a small but important class of failures.

Responsible AI requires more than explainability. Core principles include validity, safety, fairness, transparency, privacy, accountability, contestability, and human oversight. Human-in-the-loop review is meaningful only when the reviewer has authority, time, competence, and sufficient information to challenge the recommendation. A person who merely clicks “approve” is not effective oversight.

Generative AI introduces additional risks: fabricated information, confidential-data leakage, prompt injection, inconsistent outputs, copyright concerns, and unclear provenance. Appropriate controls include approved use cases, data-loss prevention, retrieval-source citation, testing, output validation, restricted tools, human approval for consequential actions, and clear records of AI involvement.

8. Regulatory and Ethical Compliance

Compliance has at least three layers. First, law and regulation may impose requirements concerning privacy, cybersecurity, financial reporting, product safety, labor, trade, customs, competition, environmental claims, and records. Second, contracts may create additional obligations. Third, professional and organizational ethics may demand conduct beyond the legal minimum.

Regulatory applicability varies by jurisdiction and sector. A global supply chain should maintain a legal-obligations register that identifies the relevant rule, affected process, data or transaction, accountable owner, evidence requirement, and review date. Legal review is essential because regulations change and similar concepts may have different definitions.

Ethical supply chain governance addresses power, vulnerability, and distributional effects. Examples include:

• an algorithm that systematically favors large suppliers and excludes small enterprises;
• surveillance of drivers or warehouse workers without proportional justification;
• sustainability claims based on incomplete or selectively reported data;
• dynamic pricing that exploits emergencies;
• pressure on suppliers to disclose data without reciprocal protection; and
• automated deactivation of a supplier without an appeal process.

Ethical analysis should ask: Who benefits? Who bears the risk? Who can challenge the decision? What alternatives were considered? Is the practice consistent with dignity, fairness, transparency, and stewardship? Compliance is a floor, not a complete ethical theory.

9. Critical Evaluation of Emerging Directions

Blockchain and Distributed Ledgers

A distributed ledger can provide a shared, tamper-evident record among parties that do not fully trust one another. Smart contracts can execute predefined actions when specified data conditions are met. Potential applications include provenance, trade documentation, certificates, and automated settlement.

However, blockchain does not make input data truthful. If a counterfeit product is entered as genuine, an immutable ledger preserves the false assertion. Governance is also required for identity, permissioning, data correction, key recovery, dispute resolution, confidentiality, and network participation. Public chains may create throughput, privacy, energy, and fee concerns; permissioned networks reduce some issues but reintroduce institutional governance. Blockchain is most justified when multiple parties need a shared record, reconciliation is costly, and a conventional database cannot provide adequate trust or neutrality.

Industry Clouds

Industry clouds combine cloud infrastructure, sector-specific data models, workflows, compliance features, and ecosystem integrations. They can accelerate implementation and standardization, but may create concentration risk, vendor lock-in, portability barriers, and dependence on a provider’s interpretation of industry requirements. Buyers should evaluate data portability, API quality, resilience evidence, subcontractors, service-level commitments, exit plans, and the boundary between configurable functionality and bespoke customization.

Sustainability Platforms

Sustainability platforms collect emissions, energy, labor, waste, water, and supplier information to support reporting and operational decisions. Their value depends on measurement boundaries, data quality, calculation methods, verification, and avoidance of double counting. A visually precise dashboard can still represent uncertain estimates. Organizations should document assumptions, activity data, emission factors, organizational boundaries, estimation methods, and assurance status. Platforms should support operational reduction, not merely disclosure.

Autonomous Ecosystems

An autonomous ecosystem is a network in which software agents sense conditions, negotiate or recommend actions, and execute transactions with limited human intervention. Examples may include dynamic freight procurement, automated replenishment, and machine-to-machine exception handling.

Autonomy should be graduated. Low-impact decisions can be automated; high-impact decisions should require approval, thresholds, explainable records, and a safe fallback. Important questions include who is liable for an agent’s action, how agents authenticate one another, how conflicting objectives are resolved, and how a human can stop the system. Interoperability without governance can spread errors rapidly across the ecosystem.

10. Future Workforce and Operating Model

Future supply chain professionals will not simply operate software. They will orchestrate data, technology, suppliers, and decisions. Important capabilities include:

• process and systems thinking;
• data literacy and statistical reasoning;
• cybersecurity and privacy awareness;
• model and AI literacy;
• contract and vendor-management skills;
• sustainability measurement and assurance;
• change leadership and user adoption;
• negotiation, communication, and ethical judgment; and
• the ability to challenge automated recommendations constructively.

Organizations should develop multidisciplinary product teams containing operations specialists, data professionals, security and privacy experts, legal advisers, and affected users. Training should be continuous because platforms, threats, regulations, and business models change. The future role of humans is not necessarily reduced by automation; it shifts toward setting objectives, handling exceptions, validating assumptions, governing trade-offs, and taking accountable decisions.

Applied Case: NorthStar Foods’ Connected Cold Chain

The following is an illustrative case.

Background

NorthStar Foods, a regional food distributor, implemented a cloud TMS connected to supplier portals, carrier telematics, warehouse sensors, and a machine-learning service that predicted temperature excursions. The company planned to automate carrier selection and customer alerts.

Problem

A security review discovered shared supplier accounts, unpatched temperature gateways, excessive access to customer addresses, and no tested recovery procedure. The prediction model performed well in normal weather but generated many false alerts during a heatwave. Managers also could not reconstruct which sensor readings had triggered customer notifications.

Analysis

The risks were interconnected. Weak identity controls could permit manipulation of sensor data. Poor lineage could make food-safety investigation difficult. Excessive alerts could cause staff to ignore genuine warnings. The model had been validated on historical conditions that did not represent the heatwave. Finally, the proposed automation could create legal and reputational consequences if customers were notified incorrectly.

Approach and Solution

NorthStar assigned accountable owners for identity, sensor data, model risk, privacy, and continuity. It replaced shared accounts with named identities and multifactor authentication, segmented gateways, established patching requirements, and required suppliers to report incidents. Customer data was minimized and access was restricted by role. The organization created a lineage record linking sensor, route, model version, alert, human review, and communication.

The model was revalidated using stress scenarios and was changed from automatic customer notification to a recommendation requiring trained review. A fallback procedure used manual temperature checks and prioritized high-risk loads. Backups and restoration were tested, and a supplier exercise simulated a compromised gateway. Contracts addressed security evidence, notification time, subcontractors, and exit assistance.

Outcome

NorthStar reduced unnecessary alerts, improved investigation speed, and demonstrated recovery capability during a later carrier outage. Automation remained in place for low-risk prioritization but not for irreversible customer or regulatory communications. The company also discovered that better data governance improved operational planning independently of cybersecurity benefits.

Learning Points

The case demonstrates that governance is socio-technical. Technical controls, contracts, data quality, model validation, human judgment, and continuity planning must work together. It also shows why a model should be governed according to the consequences of its decisions, not merely its predictive accuracy.

Challenges and Limitations

Governance programs can become bureaucratic, slow innovation, and encourage superficial compliance. Excessive controls may cause users to create shadow systems. Centralized standards may ignore local legal or operational realities. AI assurance techniques remain immature for some models, and fairness cannot always be reduced to one numerical criterion. Blockchain projects may solve coordination problems poorly addressed by ordinary databases. Sustainability data may remain estimative despite sophisticated software.

Leaders should therefore use proportional controls, involve users early, measure control outcomes, review exceptions, and retire ineffective requirements. Governance should enable responsible action rather than merely prohibit risk. Independent challenge is valuable, but accountability must remain connected to operational decisions.

Best-Practice Implementation Roadmap

A practical sequence is:

1. Inventory critical processes, data, systems, interfaces, models, suppliers, and legal obligations.
2. Classify services and decisions by operational, financial, privacy, safety, and ethical impact.
3. Assign accountable owners and establish decision rights.
4. Implement baseline IAM, logging, encryption, patching, backup, and supplier controls.
5. Map data lineage and retention requirements.
6. Define model inventory, validation, monitoring, human oversight, and retirement procedures.
7. Test incident response, continuity, recovery, and supplier escalation.
8. Integrate audit evidence into normal workflows.
9. Pilot emerging technologies against a clearly defined problem and measurable outcome.
10. Review governance after incidents, major changes, regulatory developments, and lessons from exercises.

Chapter Summary

Supply chain software governance establishes accountability for technology-enabled decisions and controls. Effective governance integrates least-privilege access, segregation of duties, cybersecurity, third-party oversight, privacy, resilience, auditability, and regulatory compliance. Identity controls must be supported by lifecycle management, monitoring, and clear ownership. Resilience requires tested recovery rather than unverified plans, while auditability depends on trustworthy logs and data lineage.

AI and analytical models require lifecycle governance because errors, bias, drift, misuse, and poor explanations can create operational and social harm. Responsible AI combines validity, fairness, transparency, privacy, security, accountability, and meaningful human oversight. Blockchain, industry clouds, sustainability platforms, and autonomous ecosystems offer real opportunities but do not eliminate the need for governance; each introduces distinctive questions about data truth, concentration, assurance, liability, and control.

The future workforce will combine supply chain expertise with data, cybersecurity, privacy, technology, sustainability, and ethical judgment. The central managerial challenge is to obtain the benefits of integration and automation while preserving resilience, contestability, trust, and human accountability.

Key Terms

Governance: The allocation of authority, accountability, decision rights, and controls.

Least privilege: Providing only the access necessary for an authorized task.

Identity and access management (IAM): Processes and technologies for identifying users and controlling permissions.

Segregation of duties: Separating incompatible responsibilities to reduce fraud and error.

Multifactor authentication: Authentication using two or more independent evidence categories.

Third-party risk: Risk arising from suppliers, service providers, contractors, or partners.

Resilience: The ability to withstand disruption, continue critical operations, and recover.

Recovery time objective (RTO): Maximum acceptable time to restore a service.

Recovery point objective (RPO): Maximum acceptable age of restored data.

Auditability: The ability to reconstruct actions, decisions, data, and controls.

Data lineage: A record of data origins, transformations, uses, and outputs.

Model risk: Risk arising from incorrect, misused, misunderstood, or poorly governed models.

Responsible AI: Development and use of AI consistent with validity, fairness, safety, transparency, privacy, and accountability.

Privacy by design: Incorporating privacy protections into systems and processes from the outset.

Pseudonymization: Replacing identifying information with substitutes while retaining a controlled possibility of re-identification.

Distributed ledger: A shared, replicated, and tamper-evident record maintained across network participants.

Industry cloud: A cloud offering with sector-specific data, workflows, integrations, and controls.

Autonomous ecosystem: An interconnected network in which software agents sense, decide, and act with limited human intervention.

Control evidence: Records demonstrating that a control exists and operates effectively.

Review Questions

Short Answer Questions

1. What is the difference between governance and management in supply chain software?
2. Why is segregation of duties important in procurement and payment processes?
3. Distinguish RTO from RPO.
4. Why does blockchain not guarantee the truth of supply chain data?
5. Identify four principles of responsible AI.

Long Answer Questions

1. Design a governance structure for a cloud-based supply chain platform involving suppliers, carriers, and internal users.
2. Explain how privacy, cybersecurity, and data lineage interact in an integrated supply chain system.
3. Describe a complete model-risk lifecycle for an inventory forecasting model.
4. Critically assess the advantages and limitations of industry clouds and sustainability platforms.
5. Explain why resilience requires both technological controls and organizational exercises.

Analytical / Application Questions

1. A company wants to automate supplier selection using an AI model. What risks, controls, human-review requirements, and audit evidence should be established before deployment?
2. A ransomware attack disables a TMS and corrupts recent shipment data. Develop a response and recovery sequence using RTO, RPO, supplier coordination, and data-integrity checks.
3. Evaluate whether a distributed ledger is justified for a multi-tier product-provenance problem. Compare it with a conventional database using trust, data quality, privacy, governance, cost, and dispute-resolution criteria.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes governance from management in supply chain software?

A. Governance executes daily transactions while management sets legal requirements.

B. Governance allocates decision rights and accountability while management operates processes.

C. Governance is limited to cybersecurity while management is limited to procurement.

D. Governance eliminates the need for operational controls.

Correct Answer: B. Governance allocates decision rights and accountability while management operates processes.

Question 2

Which control most directly reduces the risk that one employee creates a supplier and releases payment to that supplier?

A. Data compression

B. Segregation of duties

C. Load balancing

D. Network caching

Correct Answer: B. Segregation of duties

Question 3

An RPO of 15 minutes primarily specifies:

A. The maximum time allowed to restore service

B. The maximum acceptable age of restored data

C. The maximum number of recovery personnel

D. The maximum duration of a supplier contract

Correct Answer: B. The maximum acceptable age of restored data

Question 4

What is a fundamental limitation of blockchain for product provenance?

A. It cannot record events in sequence.

B. It automatically verifies the physical authenticity of every product.

C. It preserves inaccurate input data if false data is entered.

D. It cannot be used by more than one organization.

Correct Answer: C. It preserves inaccurate input data if false data is entered.

Question 5

Which practice provides the strongest evidence that backups support recovery?

A. Reporting that backup jobs completed

B. Increasing backup frequency without testing

C. Testing restoration and validating recovered data

D. Storing all backups on the same production system

Correct Answer: C. Testing restoration and validating recovered data

Question 6

Which activity belongs primarily to the second line of responsibility?

A. Operating the warehouse picking process

B. Establishing risk standards and challenging operational controls

C. Executing customer deliveries

D. Approving every routine purchase order

Correct Answer: B. Establishing risk standards and challenging operational controls

Question 7

Why can high model accuracy be insufficient for a supply chain AI system?

A. Accuracy is unrelated to data quality in every situation.

B. Accuracy may conceal poor performance for an important minority class or costly error.

C. Accurate models cannot be monitored after deployment.

D. Accuracy prevents human oversight.

Correct Answer: B. Accuracy may conceal poor performance for an important minority class or costly error.

Question 8

Which statement about pseudonymization is correct?

A. It always makes data anonymous under every privacy law.

B. It replaces identifiers but may still permit controlled re-identification.

C. It eliminates the need for access controls.

D. It guarantees that data is accurate.

Correct Answer: B. It replaces identifiers but may still permit controlled re-identification.

Question 9

Which characteristic is most important when evaluating an industry cloud provider?

A. The provider’s advertising budget

B. The color scheme of its user interface

C. Data portability, resilience evidence, and an exit plan

D. The number of unrelated consumer applications it offers

Correct Answer: C. Data portability, resilience evidence, and an exit plan

Question 10

Meaningful human oversight of an AI decision requires that the reviewer:

A. Automatically approve every recommendation

B. Have authority, competence, time, and information to challenge it

C. Be excluded from the audit trail

D. Review only the model’s user interface

Correct Answer: B. Have authority, competence, time, and information to challenge it