The Singapore Law Gazette

The 4R Decision Framework — A Lawyer’s Guide to Deciding When (NOT) to Use AI

The legal profession, long defined by its reliance on human expertise and established precedents, is undergoing a seismic shift with the rise of artificial intelligence (AI). AI-powered tools such as Microsoft Copilot, ChatGPT, Harvey, and other advanced platforms are reshaping legal workflows. They streamline legal research, automate contract analysis, and provide data driven insights that support case outcome predictions. These technologies offer law firms the promise of greater efficiency, reduced costs, and enhanced decision-making capabilities, thus enabling firms to deliver faster, more strategic services.¹How AI is reshaping the future of legal practice, Law Society of England and Wales (20 November 2024), Available: https://www.lawsociety.org.uk/topics/ai-and-lawtech/partner-content/how-ai-is-reshaping-the-future-of-legal-practice.

The appeal to utilize AI tools is undeniable. Faced with tighter budgets, higher client expectations, and growing volumes of information, law firms are under pressure to do more with less. While AI presents a compelling solution, it also raises a crucial question: Just because we can use AI, should we?

The integration of AI in workflows presents both significant opportunities and notable risks. When employed judiciously, AI can significantly boost productivity and free lawyers to focus on higher-value, strategic work. Conversely, its misuse can lead to costly mistakes, ethical breaches, and a erosion of trust in both the technology and the organisation employing it. Concerns relating to algorithmic bias, confidentiality breaches, and lack of accountability — exemplified by cases like Mata v Avianca, where AI-generated fictitious citations were submitted to court — highlight the need for careful, informed adoption.²Practical Lessons from the Attorney AI Missteps in Mata v. Avianca, Association of Corporate Counsel ( 8 August 2023), Available: https://www.acc.com/resource-library/practical-lessons-attorney-ai-missteps-mata-v-avianca. As regulators work to establish ethical guidelines, lawyers must discern when AI is appropriate and when human judgment must prevail.

To help bridge this gap, I introduce the 4Rs Framework: Repetition, Risk, Regulation, and Reviewability. Whether you are a partner responsible for developing firm-wide AI policies or an associate weighing the appropriateness use of AI for a specific task, the 4Rs provide a systematic and practical approach to leveraging AI’s benefits while managing its limitations.

1. When to Use (or Avoid) AI in Legal Work

Making informed decisions about using AI in legal practice requires an approach beyond mere reaction to industry hype or avoiding new technology out of fear. Lawyers need a clear practical framework to assess when AI truly adds value and when human judgment must take precedence.

The 4Rs evaluate the suitability of AI through four key factors. Each “R” poses a straightforward yet fundamental question: Is this a task where AI can add value, or is it one where human judgment and expertise is essential and irreplaceable?

a. Repetition — Automating the Mundane

AI excels at pattern recognition and automating repetitive tasks. When legal work involves structured inputs and predictable outputs (e.g. scanning large volumes of similar contracts, extracting key terms, or summarizing lengthy judgments) AI can significantly improve both efficiency and accuracy.

Use AI when:

• Tasks are routine and repeatable. Examples include document review, contract analysis, legal research, and e-discovery. Tools like Kira or Relativity aiR have demonstrated significant reduction in manual workload while maintaining high precision. This is especially observed in relevance review across large document sets. For instance, a study by LawGeex found that AI achieved a 94% accuracy rate in identifying risks in non-disclosure agreements (NDAs), outperforming human lawyers who averaged 85%.³LawGeex Hits 94% Accuracy in NDA Review vs 85% for Human Lawyers (26 February 2018), Available: https://www.artificiallawyer.com/2018/02/26/lawgeex-hits-94-accuracy-in-nda-review-vs-85-for-human-lawyers/.
• Managing large datasets with predictable formats. Processes such as client intake using digital forms or billing automation benefit from AI’s consistency and rules-based logic. Automating these workflows frees up legal professionals on higher-value work, strategic work.
• Drafting preliminary documents. AI can generate first drafts of standard contracts or memos, which can then be refined and contextualized by human lawyers.

Exercise caution when:

• Tasks require nuanced legal interpretation or bespoke drafting. Crafting tailored agreements or addressing novel legal questions often demands professional discretion and deep legal expertise. Large Language Models (LLMs) tend to provide only a narrow judicial perspective and can overlook broader nuances and diverse interpretations, potentially leading to representational harm.
• Reviewing complex or sensitive clauses. Context is critical for these tasks. A 2024 Stanford study found that AI misclassified 12% of complex contract clauses when not reviewed by a human. This emphasizes the continued importance of human oversight, especially in high-stakes matters.

b. Risk — Balancing Efficiency and Accountability

Not all legal tasks carry the same level of risk. Using AI to draft internal notes is fundamentally different from relying on it for client-facing advice or strategic legal decisions. When deploying AI in legal work, it is essential to consider potential worst-case scenarios, determine appropriate accountability, and clarify who is responsible if something goes awry.

Use AI when:

• The task is low-risk and/or the output is easily reversible. For example creating brainstorming outlines, drafting agenda templates for legal project teams, or generating task lists and project timelines.
• A clear review process is in place. Ensure that all AI-generated outputs are thoroughly reviewed before being shared externally or with clients.
• AI supports, but does not replace, human judgment. Use AI for predictive analytics or risk assessments that inform decision-making. For instance, use LexCheck AI to flag high-risk contract clauses (using color-coded red/yellow/green risk levels) for legal teams to review before approval.

Exercise caution when:

• The task involves high stakes. This includes providing legal advice, making strategic decisions, or submitting documents to clients or courts. Errors in these situations could result in liability, reputational damage, or breach of confidentiality.
• Handling sensitive data. Be especially cautious with generative AI tools like ChatGPT which may retain user inputs. Without robust encryption or privacy controls, these tools should not process confidential client information.
• Engaging in strategic or ethically complex decisions. Tasks such as criminal sentencing, hiring recommendations, or due diligence in sensitive matters require careful oversight. Algorithms may appear objective but can perpetuate hidden biases if not properly monitored. A 2023 study by Harvard Law School study found that certain AI tools used in hiring processes disproportionately favored candidates from elite academic institutions, raising concerns about fairness and systemic bias.⁴The AI Ethics Dilemma: How Law Firms Are Navigating the Challenges of Artificial Intelligence (19 March 2025), Available: https://vault.com/blogs/vaults-law-blog-legal-careers-and-industry-news/the-ai-ethics-dilemma-how-law-firms-are-navigating-the-challenges-of-artificial-intelligence.

c. Regulation –– Navigating Compliance Landscapes

As legal professionals, we are bound by professional conduct rules, data protection laws, and often, client-specific obligations. The use of AI in legal practice demands careful attention to these regulatory frameworks, as well as ongoing vigilance for legislative changes. Many AI tools process data on external servers which can raise concerns about confidentiality, data residency, and cross-border data transfer requirements.

Use AI When:

• The tool is compliant with applicable legal and regulatory standards. This includes bar association guidance, privacy laws such as the European Union’s General Data Protection Regulation (GDPR) and Singapore’s Personal Data Protection Act 2012 (PDPA), as well as specific client requirements. For example, AI tools used for cross-border contract review should have safeguards that align with restrictions on international data transfers.
• You are working with de-identified, anonymized, or non-confidential information. In these cases, privacy risks are minimal, making AI usage less complex from a compliance perspective.

Exercise caution when:

• You are unclear about data storage and jurisdiction. Be wary if the AI provider uses third-party cloud infrastructure and does not clearly state where and how your data is stored or protected.

• There are explicit contractual or regulatory restrictions. Avoid the use of unregulated AI tools if you are unsure about compliance, and always seek guidance when in doubt. The European Union’s Artificial Intelligence Act (AI Act) imposes strict requirements and can levy fines of up to €35 million for misuse of non-compliant AI tools. Similarly, China’s Supreme People’s Court has issued judicial guidelines emphasizing the need for human oversight and the right to appeal AI-generated decisions.
• The tool lacks recognised security certifications or controls. Avoid AI solutions that do not have SOC 2 certification, robust data encryption, or clear data residency protocols, as these are essential indicators of strong security and accountability.

d. Reviewability: Ensuring Human Oversight

AI should enhance, not replace, human judgment. An output that appears authoritative is not necessarily accurate. AI systems can hallucinate, make biased assumptions, or overgeneralize especially if trained on flawed or incomplete data. The most effective use cases are those where legal professionals can easily review, verify, and refine the AI’s output. As the complexity or opacity of the output increases, so too should the level of caution applied.

Use AI when:

• The output is easily verifiable. For example, tools like CoCounsel can assist with brief analysis, where citations and case law references can be cross-checked for accuracy. It is important to recognise that different AI models have varying degrees of accuracy and bias. For instance, while GPT-3.5 generally outperforms other models when analyzing judgments, it may show preferences — when asked who authored an opinion, GPT-3 attributed more opinions to the well-known justices like Justice Joseph Story than is historically accurate.⁵Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models (2024), Available: https://arxiv.org/abs/2401.01301.
• AI is used as an assistant, not a final authority. Treat AI as a co-pilot for drafting and idea generation, always ensuring that audit trails and version tracking are in place to monitor AI-generated contributions. Understand how the AI tool operates: know what data it was trained on, how it makes decisions, and where its limitations might impact legal outcomes.

Exercise caution when:

• The tool’s reasoning is opaque or not explainable. Avoid using “black box” systems especially in matters with high stakes. The case of Mata v. Avianca is a cautionary example where AI-generated, fictitious legal citations went undetected and were submitted in court.
• AI is relied on for legal interpretation or decision-making without human oversight. Legal professionals must stay informed about the capabilities and limitations of their AI tools and apply critical analysis to all outputs. AI models are often overconfident, particularly in complex tasks or less familiar legal areas, and may assume the truth of incorrect premises.
• The volume of AI-generated content is too large to review thoroughly. If meaningful review is not possible, the reliability and safety of using AI in that context should be reconsidered.

2. Putting It Together

AI Use Decision Flowchart

It is important to remember that AI is most effective when combined with legal expertise even when the answer to all 4Rs is “Yes”. The value of the “human-in-the-loop” principle cannot be overstated. Every dataset has a context, a source, and potential biases. If users do not question where the dataset comes from and who it might exclude, they risk perpetuating existing inequalities. Therefore, if no human is reviewing the output or if meaningful review is not possible, the use of AI should be reconsidered, regardless of the tool’s efficiency or appeal.

3. Beyond the Framework: Nuance Matters

While the 4Rs provide a practical starting point for evaluating the use of AI in legal practice, it is not a set of hard-and-fast rules. The reality of legal work is often nuanced and rarely straightforward; tasks often do not fit neatly into predefined categories. Each law firm operates within its own unique context shaped by the needs of the organization’s client base, risk tolerance, and regulatory requirements.

The purpose of the framework is to encourage thoughtful, structured analysis and not replace professional judgment or institutional values. AI related decisions often fall into gray areas where straightforward answers are elusive. For example:

• A task may begin as repetitive but evolve to require nuanced legal interpretation.
• A seemingly low-risk task may involve politically or commercially sensitive issues.
• An AI tool may appear compliant but lack sufficient transparency or governance safeguards.

In these situations, context, culture, and values are essential for making the right decision.

a. Consider the Human Impact

Even if a task meets all 4R criteria, it is important to reflect on the human element:

• Could a junior lawyer benefit from learning through manual execution of this task?
• Will over-reliance on AI decrease the team’s engagement with the material?
• Might a client perceive a diminished level of care if they subsequently discover a tool drafted their memo?

Sometimes, the convenience of AI comes at the cost of lost connection or diminished competence.

b. Set Internal Policies and Educate Lawyers

Law firms should go beyond checklists by developing comprehensive internal policies for AI use. This includes establishing protocols around data security, client confidentiality, and regular audits of AI tools to ensure compliance and effectiveness.

Equally important is fostering a culture of AI-readiness. A decision framework is only part of the solution — firms must also cultivate the right mindset and habits across the organisation. This involves:

• Providing ongoing training on legal technology and AI ethics.
• Encouraging open, transparent conversations about AI use in client matters.
• Defining and establishing clear internal policies or guidelines on approved tools, data privacy, and human review processes.

Trust in AI does not come solely from software, but on a culture of thoughtful implementation, open communication, and shared responsibility.

c. Promote a Culture of Augmented Judgment

The goal is not automation for its own sake. Law firms should cultivate a culture where AI augments human expertise, freeing lawyers to focus on complex, strategic, and client-facing work. Teams should be encouraged to scrutinize AI outputs, seek second opinions, and maintain healthy skepticism especially in high stakes matters.

AI is not meant to replace legal professionals, but to enhance their capabilities. The firms that thrive in an AI-enabled future will regard AI as their collaborator, not a driver:

• Use AI to generate, not to decide.
• Use AI to speed up, not to skip over.
• Use AI to amplify judgment, not to bypass it.

4. Judgment is the G.O.A.T

AI, algorithms, institutions, and even trusted mentors can all provide valuable insight. However, if we stop practising discernment and trade our own judgment for convenience, comfort, or consensus, we risk losing our ability to determine what is truly authentic and appropriate. When we relinquish that responsibility, someone or something else will inevitably assume it.

In a profession built on nuance, interpretation, and trust, the rise of AI does not diminish or replace the core competencies of lawyers. It simply raises the bar for how judgment must be applied.

The most effective legal professionals in the era of AI will not be those who resist change nor those who blindly adopt every new tool. Rather, they will be those who carefully discern when to lean on AI and leverage AI’s capabilities, and when to rely on their expertise.

Responsible AI integration should augment the work of lawyers, clients, and judges and not risk dehumanizing the law. The 4R Framework — Repetition, Risk, Regulation, and Reviewability — is meant to guide critical thinking, not to serve as a rigid roadmap. It leaves space for the discretion, integrity, and client focus that define truly excellent lawyering.

So, before you hit “generate”, ask yourself:

• Is this task really best suited for a machine?
• Or does this situation demand the irreplaceable human touch?

In an increasingly automated future, clear thinking, ethical judgment, and trusted relationships will remain the qualities that set exceptional legal professionals apart.