A recent social media post by a Bangladeshi judge offers a revealing glimpse into the growing role of artificial intelligence (AI) in legal practice.
Reflecting on his courtroom experience, the judge recounted that a lawyer had submitted a Naraji petition drafted with the assistance of AI.
He observed that it was perhaps the first petition he had encountered in several months that presented coherent and well-reasoned legal arguments.
When asked whether AI had been used, the lawyer responded with an embarrassed smile.
The growing acceptance of AI
The judge further noted that the increasing workload faced by lawyers often leaves them with limited time, forcing legal assistants to prepare petitions such as applications for Naraji, discharge, and bail in haste, often at the expense of quality.
The judge went on to describe AI as a “second brain” for legal professionals. While seemingly modest, this observation captures a profound shift taking place within the legal profession.
It reflects not only the growing acceptance of AI as a practical tool for legal research and drafting but also signals the technology’s expanding influence on the administration of justice.
More importantly, it foreshadows a future in which artificial intelligence may become an integral part of legal decision-making, raising important questions about the opportunities and the risks it presents for courts and the justice system.
That observation captures a broader reality. AI is no longer a futuristic technology confined to research laboratories. It has become part of everyday governance, legal practice, and public administration.
Bangladesh has already begun integrating AI into public services. AI-assisted e-traffic prosecution system ,which automatically detects traffic violations through smart cameras, is one notable example. Such initiatives promise greater efficiency, transparency, and consistency in law enforcement.
Yet technological progress also raises difficult legal questions. As AI becomes increasingly capable of generating, analyzing, and manipulating digital content, the justice system must confront an uncomfortable reality. How should courts determine what is real when technology can convincingly manufacture what appears to be truth?
This question is no longer hypothetical. Courts will inevitably encounter AI-generated audio recordings, videos, photographs, and documents.
Judges, investigators, prosecutors, and defence lawyers will increasingly be required to distinguish authentic evidence from sophisticated digital fabrications.
Establishing authenticity
The challenge is no longer simply about admitting digital evidence. It is about establishing its authenticity.
Bangladesh’s Evidence Act, 1872, amended in 2022, recognizes electronic records as admissible evidence. The amendment represented an important step towards modernizing evidentiary law.
However, it was enacted before generative AI, deepfakes, voice cloning, and large language models became mainstream technologies.
Consequently, while the Act recognizes digital records, it offers no specific legal framework for AI-generated evidence, algorithmic decision-making, or synthetic media. The result is a significant evidentiary gap.
The recently enacted Cyber Security Act, 2026 has not filled this gap either. Although the legislation addresses a range of cyber-related offences, it provides no legal definition of artificial intelligence and contains no comprehensive framework governing AI-generated evidence or the evidentiary risks posed by synthetic content.
These concerns are reinforced by the research of Dr Salwa Haque, who argues that AI should assist, not replace, human decision-making power.
AI systems learn from historical data, and where that data reflects bias, inequality, or inaccuracy, those shortcomings may be reproduced by the algorithm itself.
For that reason, AI cannot replace human judgment in evaluating credibility, weighing evidence, or appreciating the broader social and factual context of a dispute.
Perhaps the most immediate challenge comes from deepfakes and voice cloning.
Today, sophisticated AI systems can create remarkably convincing videos depicting individuals accepting bribes or produce audio recordings that faithfully imitate the voices of judges, politicians, and business leaders.
None of these events may have occurred in reality. More troubling still, current detection technologies cannot reliably identify every advanced deepfake.
Consequently, the traditional assumption that seeing is believing has become increasingly unreliable.
For centuries, courts have relied upon the principle of chain of custody to safeguard the integrity of evidence. Judges examine not only the content of evidence but also where it originated, how it was stored, and whether it remained unchanged before reaching the courtroom.
AI fundamentally complicates this process. If an entire video is artificially generated rather than merely edited, conventional forensic analysis may fail to reveal the deception.
This exposes another weakness in the existing legal framework. Section 65B of the Evidence Act requires certification of computer-generated records to establish their authenticity.
That requirement remains important, but in the age of artificial intelligence it is no longer sufficient.
Courts may also need to know which AI model generated the content, what training data it relied upon, whether the output can be independently reproduced, and whether the system has undergone independent auditing. Existing evidentiary rules do not require these safeguards.
Another significant concern is the so-called “black box problem.” Many advanced AI systems cannot explain how they reach particular conclusions. From the perspective of procedural fairness, this creates a profound dilemma.
A defendant has the constitutional right to examine and challenge the evidence presented against them. Yet meaningful cross-examination becomes difficult if neither the court nor the technology’s developer can explain how an algorithm reached its conclusion.
Equally important is the problem of algorithmic bias. AI is often perceived as objective because it relies on data rather than emotion.
In reality, AI inherits the strengths and weaknesses of the data on which it is trained. International research has repeatedly demonstrated that biased datasets can produce discriminatory outcomes, particularly in facial recognition technologies and predictive decision-making systems.
If similar technologies are adopted in Bangladesh, an obvious question arises, who bears legal responsibility when an innocent person is wrongly identified by an algorithm?
Legal reform alone will not solve these challenges. Institutional preparedness is equally critical. Bangladesh’s forensic institutions, including the criminal investigation department (CID) and the police bureau of investigation (PBI) have made significant progress in digital forensic investigations over the past decade.
However, AI-generated manipulation demands a different level of expertise. Advanced AI detection systems, machine-learning forensic techniques, metadata analysis, and independent expert evaluation will become indispensable components of future criminal investigations.
International developments offer useful guidance
Canada, despite being one of the world’s leading AI research hubs, has not yet enacted a comprehensive national AI statute. Instead, Canadian courts rely on existing privacy, human rights, consumer protection, and administrative law to resolve AI-related disputes.
The European Union has adopted a different strategy through the AI Act, introducing a risk-based regulatory framework that imposes strict obligations regarding transparency, accountability, and human oversight for high-risk AI systems.
Both approaches demonstrate an important lesson, the goal is not to prohibit artificial intelligence but to regulate it in a manner consistent with democratic values and the rule of law.
Bangladesh is currently developing a national AI policy, an encouraging development. Nevertheless, meaningful reform requires more than policy declarations.
The Evidence Act should recognize AI-generated evidence as a distinct evidentiary category. Section 65B should be modernized to require disclosures concerning AI models, training data, software versions, audit trails, and independent forensic verification where appropriate.
Simultaneously, investment in forensic laboratories and specialised training for judges, prosecutors, investigators, and lawyers should become a national priority.
The Constitution of Bangladesh guarantees equality before the law, due process, personal liberty, and the right to a fair trial. These guarantees must remain fully effective regardless of technological change.
No person should be convicted solely on the basis of an opaque algorithmic assessment that cannot be independently scrutinized or challenged.
Every major technological revolution has compelled the law to evolve. Artificial intelligence represents perhaps the most transformative challenge yet because it changes not only how information is communicated but also how evidence is created, interpreted, and evaluated.
Bangladesh should embrace artificial intelligence, but it must do so with caution. The real challenge is ensuring that technological innovation strengthens, rather than weakens, the pursuit of justice.
In the age of artificial intelligence, courts are no longer tasked only with separating truth from falsehood. Increasingly, they must distinguish reality from highly convincing digital illusion.
Technology can become one of the justice system’s most powerful allies, but it can never replace judicial independence, human reasoning, or the rule of law.
Those principles, not algorithms, must remain the ultimate guardians of justice.
Kazi Latifur Reza is Head, Department of Law, Bangladesh University.