CLA News / The Algorithm as Expert: Can AI Testify in Court? Jason Lim Jie Sheng
A doctor reviews a patient’s records and prepares an expert report for the court. It is structured, coherent, and confidently reasoned. It identifies potential departures from accepted medical practice and draws conclusions about causation. On its face, it appears no different from any other expert opinion relied upon in medical negligence litigation.
However, the report was not written by the doctor. It was generated, at least in substantial part, by an artificial intelligence (AI) system. The doctor reviewed the report, made minor amendments, and signed it off. The report is now before the court as expert evidence. This scenario is no longer speculative. Across jurisdictions, artificial intelligence is already being used to summarize medical records,[1] identify clinical inconsistencies,[2] and generate structured analyses that can, with minimal revision, resemble expert opinions.[3] In practice, the line between assistance and authorship is becoming increasingly difficult to draw.
The law, however, has not kept pace.
Expert evidence, as traditionally understood, rests on a simple yet critical premise: that the opinion is human, reasoned, and capable of being tested. In Malaysia, this premise underpins section 45 of the Malaysian Evidence Act 1950[4] and is reinforced in cases such as Foo Fio Na v Dr Soo Fook Mun & Anor[5], where the Malaysian Federal Court emphasized that expert opinions must withstand logical scrutiny.
Artificial intelligence unsettles that premise. It draws conclusions without a mind, reasons without explanation, and provides authority without accountability.
This raises another question: “if an opinion appears to be expert evidence but was generated by a machine, is it truly evidence, or is it something else?”
A Framework Built on Human Judgment
Expert evidence occupies a carefully guarded space within the law. Under section 45 of the Malaysian Evidence Act 1950, opinions are admissible only where the subject matter requires specialized knowledge. The courts have consistently emphasized that such evidence must come from a person with the requisite skill and that the opinion must be grounded in a reliable and reasoned methodology.[6]
In medical negligence litigation, this framework is shaped by the Bolam[7] principle, as refined by Bolitho[8]. Malaysian courts have adopted this approach, most notably in Foo Fio Na v Dr Soo Fook Mun & Anor, where it was made clear that a body of professional opinion is not sufficient in itself; it must also withstand logical scrutiny.
This insistence on reasoned analysis is central. An expert’s value lies not merely in the conclusion reached but, in the ability, to explain how and why it was reached.
From a practitioner’s perspective, this is what gives expert evidence its weight in court. In medical negligence matters, experts are often taken through their reports line by line during cross-examination. The credibility of an opinion often turns less on the conclusion itself than on whether the underlying reasoning can withstand sustained scrutiny.
Artificial intelligence sits uneasily within this framework.
From Assistant to Invisible Co-Author
Artificial intelligence is no longer a peripheral tool. In complex medical negligence claims, where records may span multiple lever-arch files and several years of treatment history, practitioners are increasingly turning to AI tools to organize documents, construct timelines, and identify potential inconsistencies.[9]
More advanced systems now go further, i.e., by suggesting diagnostic pathways, mapping causation, and generating structured analyses that resemble draft expert opinions.
For now, these outputs are filtered through human experts. However, as AI becomes more sophisticated, the distinction between assistance and substitution becomes increasingly blurred. A report may be formally attributed to an expert, yet be materially shaped by an algorithm.
At that point, the issue is no longer merely technical. It becomes a question of substance: whether the opinion truly remains that of the expert.
It helps to distinguish four situations. AI may be used as a tool that assists the expert, for example, by organizing records or preparing a chronology. It may serve as a source of information that the expert relies on. It may act as a substantial contributor to the expert’s reasoning. Or it may, in substance, become the source of the opinion itself. The first two situations may sit comfortably within existing practice. The last two raise the real difficulty, because the further AI moves along this spectrum, the harder it becomes to say that the opinion remains that of the expert.
Can a Machine Be an Expert?
This leads to a question that sits at the center of the title: can an AI system itself be an expert? Section 45 of the Malaysian Evidence Act 1950 is framed around the opinion of a person possessing specialized skill or knowledge. On its terms, the provision contemplates a human expert, not a machine.
The more fundamental point lies beyond the statutory language. An expert is defined not only by knowledge, but by a set of legal characteristics that an AI system does not possess: personal responsibility for the opinion, independence, a duty to assist the court, and the capacity to be cross-examined.[10] An AI system holds none of these. It owes no duty to the court, carries no responsibility, and cannot be called to account.
The practical question is therefore not whether AI can produce an opinion, but at what point an AI-assisted report ceases to be the independent opinion of the expert who signs it. That is the line the courts will need to draw.
Reliability Without Explanation
The law’s requirement that expert evidence be reasoned sits uneasily with the nature of AI. Many AI systems operate as “black boxes”[11], producing outputs without a transparent chain of reasoning. This creates a direct tension with the requirement that expert opinions must withstand logical scrutiny. A conclusion that cannot be explained cannot easily be scrutinized.
In practice, this creates a very real difficulty. If an expert is asked, under cross-examination, to justify a particular conclusion that was materially influenced by AI, how far can they go? At what point does explanation give way to speculation?
The problem is compounded by the risk of error. AI systems may generate outputs that are plausible yet incorrect. In a legal context, such errors are particularly dangerous because they may not be immediately apparent. The risk is not simply that AI may be wrong, but that it may be convincingly wrong.
Admissibility or Weight?
If AI has materially influenced an expert opinion, a further question arises: does that involvement affect the admissibility of the evidence, or only the weight the court gives it?
Where the expert can independently explain, verify, and defend the opinion, AI involvement is likely to go to weight. The court can assess the reliability of the reasoning in the ordinary way. Where the expert cannot explain the reasoning behind an AI-influenced conclusion, the concern is more fundamental. It goes to reliability, and may raise a question of admissibility under section 45 of the Malaysian Evidence Act 1950, which requires a reasoned and reliable methodology.
This distinction offers the courts a workable framework. It allows AI-assisted evidence to be received where the expert remains genuinely in control of the opinion, while withholding weight, or admissibility, where the reasoning cannot be tested.
The Problem of Accountability
Traditional expert evidence is anchored in accountability. The expert is identifiable, owes a duty to the court, and can be cross-examined.
AI disrupts this structure. Where an opinion is influenced by an AI system, responsibility becomes diffuse. The expert may not fully understand the AI’s reasoning. The developers are not parties to the litigation, and the system itself cannot be held accountable.
This raises a fundamental question: who is responsible for the opinion?
From a practical standpoint, this is not an abstract concern. In contentious matters, parties will inevitably probe the extent to which an expert has relied on external tools. Where the answer is unclear, the credibility of the evidence may be undermined.
Without clear accountability, the evidential value of such opinions becomes uncertain. This does not, however, relieve the expert of responsibility. The expert who reviews and signs a report remains professionally and legally answerable for its contents. That responsibility cannot be transferred to a tool. An expert who adopts an AI-generated conclusion without independent verification may be personally accountable for it, both to the court and under the professional standards governing expert witnesses.
A Duty to Disclose
One response to this uncertainty is disclosure. Where an expert has materially relied on AI in forming an opinion, there is a strong argument that the reliance should be disclosed.
Meaningful disclosure might cover several matters: that AI was used; the identity and nature of the system; the purpose for which it was used; the extent of the expert’s reliance on its output; whether the AI-generated material was independently verified; and, where relevant, whether the underlying prompts or outputs should be made available to the opposing party.
Disclosure of this kind connects directly to the expert’s existing duties of independence and to assist the court, rather than the party who instructed them. It also allows the opposing party and the court to test how far the opinion is truly the expert’s own.
The obligation should remain proportionate. A distinction should be drawn between the routine use of technology, which need not be disclosed, and cases where AI has materially contributed to the substance or reasoning of the opinion. Only the latter should trigger a duty to disclose.
When Cross-Examination Falls Short
Cross-examination is the primary safeguard against unreliable expert evidence. It allows counsel to probe an expert’s reasoning, assumptions, and methodology.
Where AI is involved, this safeguard is weakened. An expert may be unable to explain how a conclusion was derived, not out of evasiveness, but because the underlying process is not accessible.
AI itself, of course, cannot be cross-examined. It cannot respond to challenges, clarify ambiguities, or concede error.[12]
This creates a subtle but important shift. The court may be presented with an opinion that appears structured and authoritative, yet resists meaningful interrogation. In an adversarial system, this is a difficult position to reconcile.
This suggests that the opposing party should be entitled to explore an expert’s use of AI directly in cross-examination. Where an AI system has materially contributed to a conclusion, counsel may legitimately ask, “What was the basis on which you accepted the AI-generated conclusion?” or “Did you independently arrive at this conclusion, or did you adopt the conclusion suggested by the AI system?” Questions of this kind tie the use of AI back to the adversarial safeguards that already surround expert evidence.
Implications for Medical Negligence Litigation
In medical negligence cases, where outcomes often turn on expert evidence, these issues are particularly acute.
AI-generated outputs may begin to resemble a “body of opinion,” potentially influencing how the Bolam standard is understood. At the same time, increased reliance on AI may lead to more standardized reports, reducing the diversity of perspectives upon which courts rely to evaluate competing arguments.
That said, the potential benefits should not be ignored. AI may reduce the time and cost involved in preliminary analysis, particularly in cases where litigants struggle to obtain early expert input due to financial constraints. Obtaining expert input early is often difficult in practice. Medical negligence litigation in Malaysia is factually complex and depends heavily on scientific evidence and expert opinion, which makes it disproportionately expensive for plaintiffs, and the courts have recognized that the costs of such claims may even exceed the damages awarded.[13] Suitably qualified medical experts are also limited in number, are frequently reluctant to take on medico-legal work, and are expensive to instruct. For plaintiffs of modest means, the cost and delay of securing an early expert report can be a real barrier to pursuing an otherwise meritorious claim. Used with care, AI may help practitioners form an early view of the strength of a claim before the expense of a formal expert report is incurred.
The challenge lies in ensuring that efficiency does not come at the expense of evidential integrity.
A Commonwealth in Transition
The Malaysian position forms part of a wider Commonwealth response. Several jurisdictions have begun to issue guidance on the use of generative AI in litigation, and the direction of travel is instructive.
In England and Wales, the Courts and Tribunals Judiciary has issued guidance for judicial office holders on artificial intelligence, first published in December 2023 and updated since.[14] It warns that AI can produce plausible but inaccurate material, and reminds judges and practitioners that responsibility for what is put before the court remains human. That message was reinforced in Ayinde v London Borough of Haringey, and Al-Haroun v Qatar National Bank[15], where the Divisional Court stressed the basic duty on lawyers to check the accuracy of material placed before the court, after AI tools had produced fictitious citations.
Australia has gone further on the specific question of expert evidence. Practice Note SC Gen 23 of the Supreme Court of New South Wales, which was issued on 28 January 2025[16], provides that generative AI must not be used to draft or prepare the content of an expert report, or any part of one, without the prior leave of the court. Where leave is granted, the expert must disclose how the tool was used.
Singapore has taken a technology-neutral position. Its Guide on the Use of Generative Artificial Intelligence Tools by Court Users, effective from October 2024[17], does not prohibit AI outright, but holds court users responsible for the accuracy of AI-generated material and prohibits the use of generative AI to generate or alter evidence.
These approaches differ in detail. They share a common thread: the human user, not the machine, remains responsible for what is placed before the court. The Malaysian question is therefore part of a broader Commonwealth development, not an isolated one.
Guidance from the Malaysian Bar
Malaysian practitioners do not approach this in a vacuum. The Malaysian Bar has already issued guidance on the use of generative AI in legal practice. Circular No 342/2023 set out the risks and precautions of using generative AI, and Circular No 242/2025 updated and supplemented it.[18]
The guidance is directive. Generative AI output must be independently verified against established legal sources, treated as no more than a suggestion rather than an authority, and never used to generate legal advice, opinions, or conclusions unless every statement is independently validated. It warns specifically that these tools can produce hallucinated citations and fake cases. Above all, it reminds practitioners that the lawyer bears responsibility for the content and advice placed before the client and the court.
This guidance is addressed to advocates and solicitors, not to expert witnesses. But the principles carry across. If a lawyer may not delegate professional judgment to an AI tool, an expert whose opinion is placed before the court should be held to no lesser standard. The output remains a suggestion. The responsibility remains human.
A Principled Way Forward
The solution is not to exclude AI, but to approach its use with greater clarity and discipline.
At a minimum, experts should be expected to disclose the extent of AI involvement. More importantly, the opinion must remain that of the expert, who must be able to explain and defend it independently.
Courts may also need to scrutinize AI-assisted evidence more closely, particularly where the reasoning is unclear. Where an opinion cannot be adequately explained, it should be treated with caution, regardless of how persuasive it may appear.
For practitioners, the use of AI is not merely a matter of efficiency. It also engages professional duties of competence, independence, and candor. The comparative experience, and the Malaysian Bar’s own guidance, point in a consistent direction: disclose where AI has materially contributed, treat its output as a suggestion rather than an authority, and insist that the opinion remain one the expert can explain and defend as their own. The then Chief Justice of Malaysia made much the same point extra-judicially, observing that AI is not bound by professional ethical obligations and remains prone to hallucination, so that it can serve only as a guide and never the final determinant, and that those who use it remain answerable for it.[19]
Conclusion: Expertise at a Crossroads
Artificial intelligence does not simply introduce new tools into legal practice. It challenges the very concept of expertise upon which the legal system depends.
If expert evidence is to remain credible, it must be intelligible, testable, and accountable. AI, in its current form, sits uneasily with all three.
The question is no longer whether machines can produce answers. They clearly can. The more difficult question is whether those answers can be trusted in a legal system that depends not only on outcomes but also on the ability to understand and challenge how those outcomes are reached.
For now, the law remains anchored in human judgment. Whether it can or should adapt to a future in which machines begin to share that role is a question that Malaysian courts, and courts across the Commonwealth, will soon have to confront.
If the law cannot ask how an answer was reached, it may soon find itself accepting answers it does not fully understand.
Jason Lim Jie Sheng
Senior Associate MahWengKwai & Associates Advocates and Solicitors, Malaysia
[1] Peters, S. G., et al. (2026). Generative artificial intelligence for inpatient documentation summarization: mixed-methods quality assessment and early real-world experience. Journal of the American Medical Informatics Association, 33(8), 1466–1473.
[2] Salam, B., Stüwe, C., Nowak, S., et al. (2025). Large language models for error detection in radiology reports: a comparative analysis between closed-source and privacy-compliant open-source models. European Radiology, 35, 4549–4557.
[3] Ministeri, F., Esposito, M., Francaviglia, M., Di Mauro, L., Pantè, G. G., Salerno, M., Pomara, C., & Sessa, F. (2026). Generative artificial intelligence in forensic medicine: A pilot study on AI-simulated medico-legal reports in healthcare liability cases. International Journal of Legal Medicine
[4] Act 56
[5] [2007] 1 MLJ 593
[6] Junaidi bin Abdullah v Public Prosecutor [1993] 3 MLJ 217
[7] Bolam v Friern Hospital Management Committee [1957] 2 All ER 118
[8] Bolitho v City and Hackney Health Authority [1998] AC 232
[9] Di Mauro, L., Capasso, E., Tettamanti, C., Casella, C., Francaviglia, M., Volonnino, G., Rinaldi, R., Esposito, M., & Chisari, M. (2025). The role of artificial intelligence in analyzing clinical malpractice disputes through medical record management. Journal of Forensic and Legal Medicine, 115, 102941.
[10] John Julian van Huizen and Danial Muhamad, “Does Artificial Intelligence Require Artificial Law?” [2026] CLJU(A) lxxx
[11] Kim, C., Gadgil, S. U., & Lee, S.-I. (2026). Transparency of medical artificial intelligence systems. Nature Reviews Bioengineering, 4, 11–29.
[12] McCradden, M. D., & Stedman, I. (2024). Explaining decisions without explainability? Artificial intelligence and medicolegal accountability. Future Healthcare Journal, 11(3), 100171.
[13] Tharini Ramakrishna and Jason Low, Medical Law and Ethics in Malaysia (LexisNexis 2021) 27, paras 2-29 to 2-30.
[14] Courts and Tribunals Judiciary (England and Wales), Artificial Intelligence (AI): Judicial Guidance (first issued 12 December 2023; updated October 2025).
[15] Ayinde v London Borough of Haringey; Al-Haroun v Qatar National Bank [2025] EWHC 1383
[16] Supreme Court of New South Wales, Practice Note SC Gen 23 – Use of Generative Artificial Intelligence, commenced 3 February 2025, para 20.
[17] Supreme Court of Singapore, Guide on the Use of Generative Artificial Intelligence Tools by Court Users (2024)
[18] Malaysian Bar, Circular No 342/2023, “The Risks and Precautions in Using Generative Artificial Intelligence in the Legal Profession, Specifically ChatGPT” (24 November 2023); Malaysian Bar, Circular No 242/2025, “Updates on the Use of Generative Artificial Intelligence in Legal Practice” (3 July 2025).
[19] Tengku Maimun binti Tuan Mat, Chief Justice of Malaysia, Opening Address at the 37th LAWASIA Conference (Kuala Lumpur, 13 October 2024).
