How to ensure responsible use of AI in the courtroom

Should artificial intelligence (AI) be allowed inside the courtroom, and if so, how far should its role extend? The question has moved at an alarming speed from academic seminars to daily practice. The most contentious issue is judgment writing itself, where the spectre of AI “hallucination”, confident, plausible-sounding, entirely fabricated output, has already embarrassed courts in several jurisdictions. What was once a theoretical concern for lawyers and academics has now become a question of judicial governance.

Research increasingly shows how deeply AI has entered judicial systems. A Northwestern University study published in March 2026, based on a stratified random survey of active US federal judges, found that more than 60 per cent of respondents had used at least one AI tool in their judicial work, though mostly for legal research and document review, not for drafting judgments. This distinction matters because when properly deployed, AI can certainly assist in converting voluminous and disorganised case records into neutral, structured summaries, highlighting relevant issues, evidence, chronology, and inconsistencies without interfering with judicial reasoning. However, when used without adequate safeguards, it may undermine independent verification by encouraging reliance on algorithmic outputs that may appear authoritative despite being erroneous.

Tap here to add The Daily Star as a trusted source

According to the Law Minister’s statement in Parliament this year, nearly 4.64 million cases remained pending across all levels of the judiciary as of March 31, 2026, the overwhelming majority stuck in the lower courts. Efficient and cautious use of AI could organise records, speed up research, cut the time spent on routine administrative work, and free up judges for the actual work of adjudication.

In Bangladesh, AI tools are already in use in courtrooms and chambers across the country, but there is no practice direction governing when, how, or by whom they may be used. According to the Law Minister’s statement in Parliament this year, nearly 4.64 million cases remained pending across all levels of the judiciary as of March 31, 2026, the overwhelming majority stuck in the lower courts. Efficient and cautious use of AI could organise records, speed up research, cut the time spent on routine administrative work, and free up judges for the actual work of adjudication.

However, the drawbacks of using AI are also significant. By now, the story of lawyers citing fictitious, AI-generated judgments is a common phenomenon. A database named Damien Charlotin has logged over 2,000 such incidents worldwide as of this month, most from the United States but a growing number elsewhere, including India. In one case recorded there, the Supreme Court of India set aside a customs authority’s order and remanded it for fresh consideration after finding that the decision appeared to rely on judgments and commentary generated by AI, some of which simply did not exist.

Some jurisdictions have decided to write rules on the use of AI. The EU’s AI Act sorts systems into four risk tiers and bans the “unacceptable” tier outright. Similarly, in June 2026, the AI committee of the Supreme Court of India published draft Regulations for Use of Artificial Intelligence in Courts, 2026, built on five principles: human primacy, transparency, accountability, data protection, and judicial independence. The draft says that AI may assist with judicial work but never perform it itself. For instance, it cannot decide bail eligibility, assess witness credibility, predict recidivism or flight risk, etc. And crucially, no officer can cite a tool’s hallucination as an excuse for a wrong decision; the human stays accountable, always.

Bangladesh, on the other hand, still has no dedicated framework for AI in its courts. The draft National AI Policy 2024 gestures toward the judiciary as a “high-risk” domain but stops there; it doesn’t spell out enforceable rules for judges, lawyers, or court staff. That gap needs closing soon, before informal, unmonitored use hardens into unaccountable habit.

What would a workable framework actually look like? We would like to propose that a workable framework should contain, at the very least, the following three elements. First, a clear and public list of what is permitted and what is not. Legal research, case-record summarisation, translation, transcription, hearing scheduling, and case management could all sit on the permitted side.  Second, where AI use has materially contributed to a filing, lawyers and litigants should be required to disclose it. — and that use should be subject to verification by the lawyer and the presiding judge alike, not left to trust. Third and finally, an AI committee under the supervision of the Supreme Court could identify permissible and prohibited uses, issue practice directions for judges and advocates, certify AI tools, conduct regular technical and ethical audits, and supervise complaints arising from the district courts.

Two further safeguards should be incorporated alongside these. Data protection needs to be built in from the start. AI systems should have access only to what a given task actually requires, and judgments or other judicial material used to train or fine-tune such systems should be properly anonymised first. And judges, court staff and lawyers need basic AI-literacy training, not to instil blind trust in the technology, nor blanket suspicion of it, but informed, responsible use. In fact, none of this needs to wait for new legislation. The Supreme Court of Bangladesh can issue such practice directions on its own authority.

In conclusion, responsible use of AI for research, drafting, and administrative work could genuinely help raise disposal rates and ease the backlog if it is allowed to. But without adequate safeguards, it can just as easily become a new source of error in a system that can least afford.

Arzin Nawar is a student of BRAC University, Dr Tahura is a judge, Nusiba Hasan Ohee is a student of Bangladesh University of Professionals (BUP).