ARTICLE TEXT

In June 2023, a federal judge in New York fined two lawyers and their firm 5,000 dollars. Their offence was not dishonesty in the usual sense. They had filed a submission citing six judicial decisions that did not exist. The cases had been invented by ChatGPT, and the lawyers had not checked. Mata v Avianca became the first famous casualty of generative artificial intelligence in the courtroom. It has not been the last. This article asks what the episode teaches, how courts elsewhere have responded, and what Bangladesh should do.

The Promise

The promise of AI in legal work is real. Large language models can summarise documents, draft routine correspondence, translate, and find patterns in vast bodies of text. For a legal system with more than 4.6 million pending cases, the prospect of faster research, automated transcription and intelligent case management is attractive. Bangladesh’s e-judiciary project, though still awaiting final approval, has already introduced online cause lists, electronic bail bonds in several districts and virtual testimony by experts. AI could build on that foundation.

The Peril

The peril is equally real, and it comes in three forms. The first is fabrication. Language models generate plausible text, not verified fact. They produce citations that look right, with realistic names, years and page numbers, but which correspond to nothing. A lawyer who files such material misleads the court, and a judge who relies on it decides wrongly.

The second is confidentiality. Anything typed into a public AI tool may be stored, used to train the model, or exposed. Client instructions, draft judgments and privileged documents do not belong in a chatbot.

The third is the quiet transfer of judgment. In January 2023, a judge in Cartagena, Colombia, included ChatGPT’s answers in a ruling on a child’s medical costs, apparently to save time. The decision was legally sound, but it raised an obvious question: if a machine supplies the reasoning, who is the judge? Article 31 of Bangladesh’s Constitution promises the protection of law, and the essence of that protection is that a human being, accountable and reasoning openly, decides.

How Other Courts Have Responded

The judiciary of England and Wales issued its first guidance for judicial office holders in December 2023 and has updated it twice, most recently in October 2025. The guidance is permissive but careful. Judges may use AI for summaries and administrative tasks. They must not enter confidential information into public tools, must verify anything AI produces, and must be alert to signs that submissions before them were generated by AI, including invented citations, American spellings and legal arguments that do not match the facts. The 2025 update added warnings about deepfake evidence and hidden instructions embedded in documents.

In the United States, courts have moved case by case. After Mata v Avianca, many federal judges issued standing orders requiring lawyers to certify whether AI was used and to confirm that citations were checked by a human. Professional bodies have added AI competence to the duties of lawyers.

The Position in Bangladesh

Bangladesh has no guidance of any kind. Speaking at a summer school on AI governance in August 2026, Chief Justice Zubayer Rahman Chowdhury warned that AI cannot assess facts, evidence, law or a judge’s conscience, and called for laws and policies on the use of AI in the judiciary. Commentators have since argued that a courtroom AI policy is needed urgently. Meanwhile, lawyers and students are already using these tools, with no rules about disclosure, verification or confidentiality.

What a Bangladeshi Policy Should Say

A first policy need not be long. It should state five things. Judges and court staff may use AI for administrative and summarising tasks but not for deciding cases or drafting the reasoning of judgments. No confidential or privileged material may be entered into public AI tools. Advocates are personally responsible for every citation and factual assertion they file, and must verify AI-generated material; a fabricated citation should be treated as professional misconduct. Parties should disclose the use of AI in preparing evidence, especially audio-visual material, given the risk of deepfakes. And the Supreme Court should keep the policy under review, because the technology will change faster than any rule.

The Bar Council should mirror these rules in its professional conduct canons, and law schools should teach AI literacy alongside legal research, so that the next generation of lawyers knows both how to use these tools and when not to.

Conclusion

Technology has changed legal practice before. The photocopier, the database and email each brought efficiency and new forms of error. Generative AI is different in one respect: it can produce confident falsehood at scale. The rule of law depends on the accuracy of what courts are told and the integrity of how they decide. A short, clear policy now would cost little. The first Bangladeshi Mata v Avianca will cost far more.

Sources

1. Mata v Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), FindLaw: https://caselaw.findlaw.com/court/us-dis-crt-sd-new-yor/2335142.html
2. Courts and Tribunals Judiciary (England and Wales), Artificial Intelligence: Guidance for Judicial Office Holders (October 2025): https://www.judiciary.uk/guidance-and-resources/artificial-intelligence-ai-judicial-guidance-october-2025/
3. Verfassungsblog, the Colombian ChatGPT judgment: https://verfassungsblog.de/colombian-chatgpt/
4. Dhaka Tribune, Chief Justice warns against use of AI in court (29 August 2026): https://www.dhakatribune.com/bangladesh/court/418574/chief-justice-warns-against-use-of-ai-in-court
5. The Daily Star, Bangladesh needs a courtroom AI policy, and soon (30 August 2026): https://www.thedailystar.net/opinion/views/news/bangladesh-needs-courtroom-ai-policy-and-soon-4259816