Opinion: Will artificial intelligence replace lawyers?

THIS question is often asked in seminars, webinars, conferences, and quiet conversations among lawyers: Will artificial intelligence (AI) replace lawyers?

AI will not replace lawyers who think, verify, and act responsibly. But it might replace or expose lawyers who refuse to adapt, or who outsource their thinking to a machine.

It is engineered software, trained on data, and shaped by human design choices. Its output depends on the data, model, prompt, and user’s judgement. Lawyers must ask the right questions before using AI: who is using it, for what task, with what data, under what safeguards, and with what verification?

AI is a powerful tool. Its ability to search, synthesise, and simulate reasoning is improving rapidly. It is used in legal research, contract analysis, predictive analytics, and decision-support tools. It is real, and it is here to stay.

The right question is not: Can AI do what lawyers do? It is: What can only a thinking, reasoning, accountable human lawyer do? The danger is not that AI will think like lawyers. The danger is that lawyers will stop thinking like lawyers.

A lawyer is not a retrieval system, a document generator, or a pattern-matching engine running on legal training data. A lawyer thinks, reasons, and exercises judgement under genuine uncertainty and real consequences. A lawyer is responsible, accountable, and liable, to the client, the court, and the profession, for advice given and not given, and for work done and not done.

AI can produce output that resembles reasoning and sounds authoritative. But it has no conception of truth. It cannot be held professionally responsible, reprimanded, fined, suspended, or struck off. It cannot look a frightened client in the eye and say: I take responsibility for this advice.

That accountability, irreducibly human and irreducibly personal, is not a weakness of our profession. It is its entire strength and foundation.

The real threat is not that AI will replace us. But that we will let it; that we will hand over our reasoning to a tool, and call it efficiency. That is not adoption of technology. It is abdication of professional responsibility.

Our chief justice said it with striking clarity at the recent Commonwealth Legal Education Association Conference. Cautioning particularly against overreliance among young lawyers, he said: "To be human is to learn, to err, to struggle, and to remember. There is no doubt that it has made us use less of our mental faculties. When we outsource our thinking, we are not just saving time. We are losing a part of ourselves. Efficiency is for machines."

He made the same point at last year’s Malaysia Legal Forum, likening AI to a chainsaw — useful in the right hands, but super dangerous in the wrong ones.

Abdication does not always announce itself dramatically. It creeps in through small acts of uncritical reliance: copying an AI-generated summary into an advice letter without reading the judgement; filing documents built around AI-generated case citations without verification; telling a client, “AI said it,” as though that settles the matter.

The pattern is now familiar. In the United States, Mata v Avianca involved fictitious ChatGPT-generated cases and resulted in sanctions. In the United Kingdom, Ayinde v London Borough of Haringey involved fabricated authorities and a wasted costs order. In Singapore, lawyers in separate matters faced costs consequences and court scrutiny after citing non-existent AI-generated cases. Different countries, but the failure is the same: AI output was accepted without independent verification.

The Malaysian Bar moved early. In November 2023, the Bar Council issued Circular No. 342/2023 — its first formal advisory to members of the Malaysian Bar, on risks and precautions in using generative AI. The circular identified nine risk categories: hallucinated citations, bias, lack of accuracy, risks to client confidentiality and intellectual property, data privacy breaches, unpredictable behaviour, security risks including data poisoning, limited emotional intelligence, and AI hallucinations. Its closing reminder remains relevant: "ChatGPT and other GenAI products should be treated only as guidance tools. As an Advocate and Solicitor, you must apply your mind to the questions, answers, problems, or solutions. You must carefully draft your documents. Ultimately, as the appointed lawyer, you bear responsibility for the content and advice rendered to your client."

In July 2025, the Bar Council issued Circular No. 242/2025, a substantially expanded update. By then, GenAI had moved from novelty to infrastructure, and international misconduct cases had multiplied. The update referenced the 4R Decision Framework, published in the Singapore Law Gazette in May 2025, a practical tool asking four questions before using AI: Is the task repetitive and routine? What is the risk if the AI output is wrong? What regulation governs the task? Is the output reviewable and auditable?

The Bar Council distinguished recommended uses of AI — early drafts, brainstorming, checklists, formatting, and client-facing educational content — from uses to avoid — generating legal advice, verifying case law, drafting pleadings or submissions without full independent review, or handling client information and personal data.

The circular concluded: "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 when to rely on their own expertise."

A lawyer who causes harm through unverified AI-assisted work today cannot claim ignorance of the risks. At this point, ignorance is a choice; in law, wilful blindness is no defence.

The Securities Commission Malaysia (SC) published an internal AI governance framework that every lawyer should study. Plain-language. Practical. Communicated on posters within its offices. Its headline: Responsible AI equals trusted outcomes. The SC’s framework is built around accountability, transparency, reliability, privacy, security, fairness, human-centred use, and risk assessment. Its message is simple: classify risk, approve use cases, keep a human in the loop, and verify before sign-off.

Its Dos and Don'ts are direct: do not solely rely on AI outputs — they can be inaccurate, producing wrong facts and made-up references. Do not allow AI to decide on your behalf. Always apply human judgement before acting.

Beneath the practical questions of governance and verification is a deeper issue — one that will shape how AI is regulated and how professional responsibility evolves.

As AI advances — voice bots, chatbots, humanoid robots, electronic agents, and agentic systems — it will become more embedded in our lives. There will be growing pressure to treat AI as more than a tool: to speak of it as though it decides, creates, or advises; to give it a form of agency.

We should be extremely cautious about this.

These systems do not exist in a vacuum. Human beings designed, built, trained, deployed, and released them. When they go wrong, the question must be asked: who is accountable? You cannot hold the machine liable. You can shut it down or deactivate it, but the harm would already be done. From whom do you claim just recompense?

Liability, responsibility, and accountability must be ascribed to human actors and operators: the designer, developer, or deployer who uses the tool without adequate oversight and insight, or the lawyer who filed the document without reading it. These are human failures. They must attract human consequences.

If we ascribe legal personality to AI — treating it as an independent agent capable of responsibility — we would do something more dangerous than any hallucination. We would abdicate and outsource our thinking to the machine, making ourselves redundant not because AI replaced us, but because we surrendered what makes us irreplaceable.

The person who deploys AI should be liable for its acts, even those not specifically anticipated. And where governance, training, or due diligence is absent, someone should face personal liability for serious AI-related failures. You cannot deploy AI and walk away from responsibility for what it does.

Amid the excitement, anxiety, and headlines about AI, remember that we already possess the most powerful intelligence available. Every lawyer in this country has it. It requires no subscription, no data centre, no terms of use, no prompt engineering.

It is Actual Intelligence: human brainpower, designed by nature, developed by nurture, deployed by the human being. Let us use it. It is free.

This is not an argument against AI. It is an argument for perspective. Use it. Learn it. But never mistake it for a substitute for the human mind that reads the room, weighs the evidence, makes the judgement call, and stands behind the outcome.

Lawyers who think — who reason under pressure, carry real accountability, and stand behind their advice with their name and practising certificate — will not be replaced. AI demands that they become more deliberate, rigorous, honest about what they know and do not know, and courageous in engaging with powerful, unsettling tools.

The Malaysian Bar has done its part: two circulars, clear guidance, practical frameworks, issued early and updated. The judiciary has spoken. The regulatory conversation is taking shape. There is no shortage of case law or directional roadmaps.

AI assistance is acceptable. Intellectual outsourcing is not. The future belongs not to Artificial Intelligence alone, but to lawyers who keep their Actual Intelligence fully engaged.

Sathish Mavath Ramachandran is a partner at the law firm of Deol & Gill. He was chairman of the Bar Council’s Legal Tech, AI and Sandbox Committee, and the Personal Data Protection Committee. He is a Member of the Advocates and Solicitors Disciplinary Board.