Three Lines Model and AI Risk Responsibilities

Why This Topic Matters

Artificial intelligence introduces risks that cut across an entire organization: biased outputs, opaque decision logic, data leakage, model drift, regulatory exposure, and reputational harm. When something goes wrong with an AI system, the first question regulators, boards, and customers ask is simple: who was responsible? If that question cannot be answered clearly, the organization has a governance failure, not just a technical failure.

The Three Lines Model, published by The Institute of Internal Auditors (IIA) and updated in 2020, is the most widely adopted structure for assigning risk responsibilities. ISACA's AAIR (Advanced in AI Risk) body of knowledge integrates this model into the AI Risk Governance and Framework Integration domain because it provides the organizational scaffolding on which frameworks like the NIST AI Risk Management Framework, ISO/IEC 42001, and COBIT can actually operate. A framework tells you what controls and activities are needed; the Three Lines Model tells you who performs them, who oversees them, and who independently verifies them.

For the exam, this topic is heavily tested because it is scenario-friendly. Examiners can describe a situation and ask you to identify which line should act, where accountability sits, or whether an arrangement compromises independence. Mastering the model means you can reason through almost any AI governance scenario.

What the Three Lines Model Is

The Three Lines Model describes how an organization structures roles to achieve its objectives while managing risk. It replaced the older Three Lines of Defense model, and the terminology shift from defense to lines is deliberate: the updated model emphasizes value creation and protection, not just defensive control. The model identifies the following roles:

1. The Governing Body (Board or equivalent)
The governing body is accountable to stakeholders for organizational oversight. It sets the tone at the top, defines risk appetite, approves strategy, ensures resources are adequate, and establishes an independent internal audit function. For AI, the governing body approves the AI strategy and AI risk appetite, demands transparency on high-impact AI systems, and holds executive management accountable for responsible AI outcomes.

2. Management: First Line Roles
First line roles are directly aligned with delivering products and services to clients. They own and manage risk as an inherent part of operations. In an AI context, first line roles include data scientists, machine learning engineers, product owners, business unit leaders deploying AI tools, procurement teams acquiring third-party AI, and operations staff who use AI outputs. They design models, select training data, implement controls, monitor performance, and are the first to detect and correct problems. Crucially, the first line owns the risk.

3. Management: Second Line Roles
Second line roles provide expertise, support, monitoring, and challenge on risk-related matters. They help the first line manage risk effectively but do not own the operational risk themselves. For AI, second line roles typically include enterprise risk management, model risk management or model validation teams, compliance, privacy, information security, legal, and increasingly a dedicated AI governance office, responsible AI function, or AI ethics committee. They establish policies and standards, define AI risk taxonomies, maintain AI inventories, perform independent model validation, monitor key risk indicators, and escalate concerns to senior management and the governing body.

4. Internal Audit: Third Line Role
Internal audit provides independent and objective assurance and advice on the adequacy and effectiveness of governance and risk management, including the work of the first and second lines. Internal audit reports functionally to the governing body, typically through an audit committee, and must remain free of management responsibilities that could impair its objectivity. For AI, internal audit assesses whether the AI governance framework is designed appropriately and operating effectively, whether model risk controls are followed, whether the AI inventory is complete, and whether the second line is performing its challenge role credibly.

5. External Assurance Providers
Outside the three lines sit external auditors, regulators, certification bodies (for example ISO/IEC 42001 certification auditors), and independent algorithmic auditors. They provide additional assurance to stakeholders but are not part of the organization's internal structure.

The Six Principles of the Updated Model

The 2020 IIA model rests on six principles that you should be able to recognize:

1. Governance requires structures and processes that enable accountability, action, and assurance.
2. Governing body roles include integrity, leadership, transparency, setting risk appetite, and establishing internal audit.
3. Management roles span both first and second line roles; management is responsible for achieving objectives through risk-based decision-making.
4. Third line roles provide independent assurance through competent and objective internal audit.
5. Third line independence is essential and is protected by reporting lines, unrestricted access, and freedom from management responsibilities.
6. Creating and protecting value requires all roles to align, communicate, and collaborate; the lines are not silos.

How It Works in Practice for AI Risk

Consider the lifecycle of an AI system and how responsibilities flow:

Design and development. The first line defines the use case, sources data, builds the model, documents assumptions, and performs initial testing for accuracy, fairness, and robustness. The second line sets the standards the first line must meet (for example a model development standard, data quality requirements, bias testing thresholds, and documentation templates such as model cards) and reviews or validates the model before deployment. Internal audit does not participate in building or approving the model; doing so would impair its later ability to audit it.

Deployment and approval. High-risk AI systems typically pass through a governance gate. The first line submits the system with evidence; the second line (model risk management or the AI governance committee) challenges it and recommends approval, conditional approval, or rejection; senior management approves within the risk appetite set by the governing body. The governing body may require notification or approval for systems above a defined impact threshold.

Operation and monitoring. The first line monitors model performance, drift, data pipeline integrity, and incidents. The second line aggregates metrics across the AI portfolio, monitors key risk indicators against appetite, tracks regulatory change, and reports to management and the board. Internal audit periodically tests whether monitoring is actually occurring and whether exceptions are handled as policy requires.

Incidents and remediation. The first line responds and remediates. The second line ensures root causes are identified, policies are updated, and lessons are shared. Internal audit may perform a post-incident review and assess whether management's corrective actions are adequate. The governing body receives reporting on significant incidents.

Key Distinctions the Exam Expects You to Know

Ownership versus oversight versus assurance. The first line owns and manages risk. The second line oversees, supports, and challenges. The third line independently assures. If a question asks who is accountable for an AI model producing biased results, the answer is the first line business owner, not the data governance team that set the policy and not internal audit that reviewed the program.

Lines versus roles. The updated model stresses that these are roles, not rigid organizational units. One individual or team may perform both first and second line roles, though this should be managed to preserve objectivity. Only the third line requires structural independence.

Independence of internal audit. Internal audit must not design controls, own policies, make management decisions, or approve AI deployments. It may advise, consult, and recommend, but if it takes on management responsibilities it loses the ability to provide objective assurance over them. Any scenario where internal audit is asked to run the AI governance program, approve models, or serve as the AI ethics committee's decision-maker signals an independence conflict.

Second line is not a substitute for first line accountability. A common governance failure is the first line assuming the second line is responsible for compliance. Policy-setting and validation do not transfer ownership. The business still owns the outcome.

Governing body accountability cannot be delegated. The board may delegate tasks to management, but it remains ultimately accountable to stakeholders for AI oversight. Setting AI risk appetite and ensuring responsible AI governance exists are board-level responsibilities.

Coordination and combined assurance. The model encourages the three lines to coordinate so that assurance is not duplicated and gaps are not left uncovered. An assurance map showing which AI risks are covered by which line is a practical artifact that exam questions may reference.

Integration with AI Frameworks

The Three Lines Model operationalizes other frameworks:

- NIST AI RMF: The Govern function describes accountability structures, roles, and culture. The Map, Measure, and Manage functions are largely executed by the first line with second line standards and oversight. Internal audit assesses the whole.
- ISO/IEC 42001: The AI management system requires top management leadership and commitment (governing body and executive management), defined roles and responsibilities (first and second line), and internal audit of the management system (third line).
- COBIT: COBIT's distinction between governance (evaluate, direct, monitor) and management (plan, build, run, monitor) maps naturally onto the governing body and management roles, with assurance objectives supporting the third line.
- EU AI Act and similar regulation: Provider and deployer obligations fall on the first line; compliance monitoring and conformity support fall on the second line; independent review falls on the third line and external assurance.

Common AI-Specific Challenges

- Blurred lines in data science teams. A central data science center of excellence may both build models (first line) and set modeling standards (second line). The organization must recognize this and introduce independent validation to preserve challenge.
- Shadow AI. Business units adopting generative AI tools without registration undermine the second line's ability to maintain a complete inventory. The first line is responsible for registering systems; the second line for the inventory process; the third line for testing completeness.
- Third-party AI. When AI is procured, the first line still owns the risk of using it. The second line sets vendor due diligence requirements; the third line assesses whether due diligence is performed.
- Skills gaps in internal audit. Internal audit may lack AI expertise. Acceptable responses include training, co-sourcing with specialists, or using external assurance, but never transferring the assurance role to the second line.
- AI ethics committees. These are usually second line or governance advisory bodies. If a committee has decision authority over deployments, it is exercising a management role and is not independent assurance.

Exam Tips: Answering Questions on Three Lines Model and AI Risk Responsibilities

Tip 1: Identify the verb. Scenario questions hinge on action words. Owns, implements, operates, builds, monitors day-to-day, remediates point to the first line. Sets policy, defines standards, validates, challenges, monitors aggregate risk, reports to management, advises point to the second line. Independently assures, evaluates effectiveness, tests, reports to the audit committee point to the third line. Sets risk appetite, approves strategy, holds management accountable, ensures internal audit exists point to the governing body.

Tip 2: Default accountability to the first line. When asked who is accountable or responsible for an AI system's risk, the business or product owner who deploys it is almost always the correct answer. Distractors will name the model risk team, the AI ethics office, the CISO, or internal audit.

Tip 3: Watch for independence violations. If an option has internal audit approving models, writing AI policy, chairing the governance committee with voting rights, or remediating findings, eliminate it. Internal audit may recommend and advise, but not decide or own.

Tip 4: Second line challenges but does not own. If a scenario describes a second line team being blamed for a model failure, the best answer usually clarifies that the first line owns the risk while the second line should strengthen its challenge and monitoring.

Tip 5: Governing body questions test non-delegable duties. If asked what the board must do regarding AI, favor answers about setting AI risk appetite, approving AI strategy, overseeing high-impact systems, ensuring adequate resources and competencies, and establishing independent assurance. Avoid answers that have the board performing operational tasks like approving every model or reviewing training data.

Tip 6: Remember it is roles, not org charts. Questions may describe a single team doing both first and second line work. The best response acknowledges this is permissible under the updated model but requires safeguards for objectivity, not that it is automatically prohibited.

Tip 7: Look for the most structural answer. ISACA exams favor answers that address root causes and governance structures over point fixes. Establishing clear role definitions, a RACI, or an assurance map typically beats conducting more testing or training alone.

Tip 8: Connect to the frameworks. If a question references NIST AI RMF, ISO/IEC 42001, or COBIT alongside the three lines, map the governance elements of those frameworks to the governing body and management roles, and the audit or assurance elements to the third line.

Tip 9: External assurance supplements, it does not replace. Hiring an external algorithmic auditor does not eliminate the need for internal audit or the second line. Answers suggesting external review can substitute for internal roles are usually wrong.

Tip 10: Read for the question actually being asked. Distinguish between who performs an activity, who is accountable for it, and who provides assurance over it. The same AI control can involve all three lines in different capacities, and the wrong answer often describes the right activity attributed to the wrong role.

Quick Reference Summary

- Governing body: Accountable for oversight; sets AI risk appetite; ensures independent audit exists.
- First line: Owns and manages AI risk; builds, deploys, operates, monitors, remediates.
- Second line: Sets AI policies and standards; validates and challenges; monitors aggregate risk; reports.
- Third line: Independent assurance over governance and risk management; reports to the audit committee; no management responsibilities.
- External assurance: Supplementary confidence for stakeholders and regulators.

If you can place any AI-related activity into one of these roles and explain why that placement protects both accountability and independence, you are prepared for this section of the AAIR exam.