The Singapore Law Gazette

When Clients Use AI: Managing Expectations, Misinformation, and Costs

Abstract

1. Corporate clients often provide their lawyers with templates, case law or legal analysis produced using generative AI tools, which usually contain errors including incorrect citations, made up authorities or assumptions that do not align with Singapore law. This article is a practical guide for practising lawyers who are increasingly being asked to review and amend these AI generated materials. We focus on (i) how lawyers should frame their language when pushing back on flawed AI-generated content without undermining client confidence/ commercial relationships; and (ii) what lawyers can proactively put in place (contractually, procedurally and culturally) to manage their clients’ use of AI in these workstreams, instead of reacting to these issues after the fact.
2. Fundamentally, this tension reflects a simple truth: lawyers may guide and warn, but a client exercising their free will over how they use AI (while recognising the limits of this control) should be seen as an invitation to dynamic, not defensive, lawyering.

Introduction

3. Imagine this scenario: you are advising a corporate client who shares a draft agreement with you that immediately raises some red flags. The drafting is oddly phrased, key concepts are misapplied and the draft cites “Singapore law” that does not exist. Upon further discussion the client explains that the agreement was generated using OpenAI’s large language model, ChatGPT, and shares that they have been relying on AI tools for first-cut drafting and application of legal analysis.
4. This is no longer a hypothetical problem, rather a recurring issue in practice today. This article therefore sets out tips on how to respond when clients provide their lawyers with AI generated materials, without embarrassing the client, and/or being at risk of breaching one’s professional obligations.
5. The objective is not to discourage initiative, but to recalibrate expectations: to help clients understand where AI can assist and where legal judgment and jurisdiction specific expertise remain indispensable.

When do clients use AI and how should lawyers respond?

6. Clients typically turn to AI to find solutions to the following predictable pressure points: quick turnaround time, low cost and uncertainty about how to begin. Furthermore, a client’s use of AI tends to surface in these three recurring scenarios: (a) AI-generated agreements, (b) AI-generated case law citations and (c) AI-generated legal analysis. We will now unpack how each scenario can be dealt with a calibrated yet firm response.

AI-Generated Agreements

7. Clients often use AI to generate template agreements because they lack drafting experience and/ or want a fast starting point. However, AI-generated agreements frequently fail to reflect mandatory Singapore law requirements and/or import foreign law concepts without an understanding of the nuance and context attached to a matter. Common examples include not referring to relevant sections of the Companies Act 1967 and boilerplate clauses for a typical commercial agreement. In these situations, the lawyer’s first task is not correction, but positioning. A dismissive response risks alienating the client while complete acceptance of the same risks negligence and/ or recklessness on the lawyer’s part.
8. A practical approach is to acknowledge the client’s initiative and explain neutrally why AI tools often miss jurisdiction specific requirements. A response along the following lines may be effective: “Thanks for sharing this. AI-generated agreements typically require a clause-by-clause review and possibly restructuring to ensure compliance with Singapore law. We will conduct a full review rather than a high level review and shall reflect the time required in our fees.” This approach is likely to preserve the relationship while providing professional value.

AI-Generated Case Law Citations

9. Current publicly available AI tools are particularly unreliable when citing case law. They may confidently generate authorities that do not exist via “hallucinations”, misstate holdings and/or conflate legal principles across jurisdictions. A growing database seems to have recorded over 800 hallucinated case citations across the United States and United Kingdom, underscoring that this is likely a systemic limitation rather than an isolated error.
10. Clients typically rely on AI-generated case law to justify aggressive commercial positions. Here, the lawyer may quickly determine that the cited cases cannot be located on trusted databases such as LawNet. As such, a response in this situation could look like: “Before relying on the authorities cited here, we will need to independently verify each case and assess whether it applies under Singapore law. This verification exercise sits outside the original drafting scope, so let us know if you would like us to proceed on this basis.” This is likely to keep the discussion constructive while restoring to emphasising accuracy.

AI-Generated Legal Analysis

11. Clients may also rely on AI to conduct preliminary legal analysis, particularly when their matter has many moving parts and is complex/technical. However, AI tools typically run the risk of overlooking non statutory requirements such as sector specific guidelines from the Monetary Authority of Singapore or Personal Data Protection Commission for example. This demonstrates the reality that current large language models (LLM) operate through pattern matching with incomplete context, rather than legal judgment, and errors can compound rather than self-correct.
12. A constructive approach is to acknowledge the preparatory effort and a practical response might be: “We’re happy to use this as background context. Any advice we provide will be based on our independent analysis, including reference to local laws and guidelines. The deliverable will therefore be our own assessment, rather than a revision of the AI-generated analysis.”

Managing Expectations

13. Another angle emerging from client use of AI is managing the scope of services and cost of a matter. AI-generated drafts can create the false impression that legal work is largely complete, when in reality significant professional time is required to verify, correct and contextualise flawed inputs. As such, we suggest:

Refining the Engagement Letter

14. Where clients rely heavily on AI, lawyers could consider making adjustments to their engagement letters and their offered scope of services. Even light touch clarifications can help manage expectations and miscommunication. For example, engagement letters may expressly state that:
    • AI-generated materials are treated as unverified inputs and are not considered work product;
    • Additional time and costs may be required to review, correct or replace AI-generated content; and
    • Fees are assessed based on the work required to deliver accurate, Singapore law compliant advice, regardless of how much preliminary material the client has prepared.

Incorporating Checklists

15. Boundary setting works best when it is operational, not abstract. Lawyers may find it useful to adopt standardised language or internal checklists when responding to AI-generated client materials. Examples of boundary setting language include:
    • “I’m happy to review this but it will require a full verification exercise at an additional cost as AI-generated content may contain inaccuracies.”
    • “I’ll treat this as background only and may not refer to it substantially before reverting with the deliverables.”
    • “This falls outside the original scope of work. Please confirm if you would like us to factor in the additional time and fees for reviewing this before we proceed.”
16. Internally, firms may also benefit from a checklist outlining:
    • Has the AI content been verified and how long will that exercise take?
    • Does it rely on foreign or non-existent law?
    • Does correcting it require fresh legal analysis rather than editing?

Responding Constructively

17. The goal is not to discourage clients from using AI, but to recalibrate how its outputs are treated within the engagement letter. Lawyers may benefit from being explicit that correcting AI errors often expands rather than reduces scope. A constructive framing is to explain that:
    • AI may accelerate ideation, but legal responsibility remains human;
    • Verification is value-adding work, not duplication; and
    • Costs reflect time spent verifying and is not arbitrary.

Conclusion

18. At its core, client use of AI is not a challenge to legal authority but a signal of changing expectations around speed, accessibility and problem-solving. AI remains an emerging and inherently unreliable technology in the legal context given that it is in its nascent stages. Its outputs are shaped by incomplete context and probabilistic pattern matching and errors, particularly legal ones, which run the risk of compounding rather than cancelling out.
19. For lawyers, the appropriate response is not resistance, but upskilling and recalibration. AI can be a useful assistive tool, but ultimate responsibility for legal advice and accuracy remains with the lawyer, regardless of whether AI was used in the process.
20. Clients may choose to rely on AI in their own workflows and that choice ultimately lies beyond the lawyer’s control. What remain within the lawyer’s control are how expectations are framed and how professional judgment is asserted. In that sense, recognising both client autonomy and the limits of technological developments is now an integral part of the modern lawyer’s toolkit.