How Not to Use Digital Evidence in Atrocity Trials: A Cautionary Tale from the International Crimes Tribunal in Bangladesh

1. INTRODUCTION

Around the world, atrocity investigations are increasingly built on footage captured by people on their phones. The July-August Mass Uprising in Bangladesh, one of the deadliest protest the world witnessed in 2024 and often described to be the world’s first “successful” Gen Z revolution, offers a stark example of this growing phenomenon of crowd-sourced digital evidence playing a central role in human rights investigations and accountability processes in atrocity trials. According to a fact-finding report by the Office of the United Nations High Commissioner for Human Rights, in the space of three weeks, at least 1,400 people were killed and thousands more suffered life altering injuries. It became one of the most intensely recorded episodes of state-sanctioned violence in Bangladesh’s recent history, leaving behind an enormous trove of videos, images, and other digital evidence of the killings and injuries, filmed by eyewitnesses and victims themselves, that took place across the country in July and August 2024. These now form a wide and important corpus of digital evidence that is actively being used in Bangladeshi courts.

But exactly what happens when those pixels move from the streets to the courtrooms? Upon first glance, it appears to be a success story. For the first time in history, gross violations of human rights are recorded in real-time by protesters, bystanders, and journalists with phones in their hands, which can then be used as evidence to hold alleged perpetrators legally accountable. However, the trial and sentencing of former Prime Minister Sheikh Hasina Wazed by the International Crimes Tribunal (ICT) in Bangladesh shows the procedural hazards of the digital turn. 

Although Hasina is facing at least two cases before the ICT, this brief focuses specifically on the one pertaining to crimes against humanity (Case No. 02 of 2025) in which the tribunal sentenced her to death in a publicly broadcast verdict that was delivered on 17 November 2025. The trial relied heavily on videos, audio files, and forensic reports, yet it unfolded in a setting where the accused was tried in absentia, defended only by a state-appointed counsel, and ultimately sentenced to death. The result is a troubling paradox. Digital evidence was central to exposing atrocities during the Monsoon Revolution, yet it was also used in ways that risk undermining the very fair trial guarantees that international criminal law is supposed to protect and digital evidence is meant to offset. 

Of note, however, these challenges are not unique to Bangladesh. Recent efforts to use social media videos to document atrocities in places such as Syria and Ukraine, and more recently in Iran, have raised similar questions about how to guard against manipulation, misinterpretation, and over-reliance on uncorroborated online content.

It is equally important to acknowledge that there is no inherent tension between fair trial guarantees and the use of digital evidence. On the contrary, when properly collected, preserved, verified, and tested, digital evidence can strengthen the integrity and robustness of fair trial processes. The real question, therefore, is not whether digital evidence should be used in atrocity cases — it clearly should. Instead, the question is how courts, defence counsel, and prosecutors can leverage digital evidence in a way that does not erode fair trial guarantees.

This policy brief uses the handling of digital evidence by the ICT in the trial of Hasina as a case study to expose three structural vulnerabilities of the digital turn in atrocity prosecutions: shallow tests of authenticity, fragmented chains of custody, and systematic inequality of arms. These vulnerabilities are assessed against the fair trial standards reflected in Bangladesh’s Constitution, the International Crimes (Tribunals) Act 1973 (the 1973 Act), Article 14 of the International Covenant on Civil and Political Rights (ICCPR), and the related protections governing criminal trials and capital sentencing. It argues that unless courts confront these vulnerabilities directly, the mishandling of digital evidence will risk diluting the minimum fair trial guarantees under domestic constitution, international human rights law and international criminal law. 

While the Hasina trial involves other procedural and substantive issues, the analysis in this policy brief is strictly confined to these three specific vulnerabilities. Similarly, the examples drawn from the proceedings are utilised solely to illustrate these distinct evidentiary challenges, acknowledging that they may also invariably touch upon broader legal concerns that fall outside this brief’s narrowly defined scope. Additionally, digital evidence databases — seeking to systematically collect, preserve, verify, and organise open-source and other forms of digital evidence — also carry inherent limitations, including risks of selective documentation, platform dependence, and uneven access to evidence; these issues likewise fall outside the scope of this paper. Ultimately, through its focused examination of these three evidentiary vulnerabilities, this brief aims to set out practical steps for improving the use of digital evidence in atrocity trials, with the objective of strengthening accountability while safeguarding fair trial guarantees.

2. LEGAL AND POLICY FRAMEWORK ON USE OF DIGITAL EVIDENCE IN ATROCITY TRIAL

Bangladesh’s framework for atrocity trials combines constitutional carve-outs, statutory fair trial guarantees, international human rights obligations, and a flexible evidentiary regime governing the admission and assessment of digital evidence.

Although the Constitution of Bangladesh guarantees a list of judicially enforceable fundamental rights, Article 47A provides that those accused of committing genocide, crimes against humanity, and related international crimes cannot invoke certain fundamental rights. Specifically, these include Article 31 ensuring the right to protection of law, and the first and third clauses of Article 35 safeguarding against non-retroactivity and providing the right to a speedy and public trial by an independent and impartial judiciary, respectively. Going further, it also bars persons accused of the offences from appealing to the Supreme Court of Bangladesh for constitutional remedies. These carve-outs have attracted criticisms because they lower the domestic baseline of fair trial protections precisely in cases where the stakes are highest. 

At the same time, the 1973 Act, as amended, creates an express statutory obligation in section 6(2A) that the ICT “shall be independent in the exercise of its judicial functions and shall ensure fair trial.” Additionally, section 17 of the 1973 Act enumerates specific rights of the defendant, including the right to conduct one’s own defence or be assisted by counsel, to present evidence, to cross-examine prosecution witnesses, and to have adequate time and facilities to prepare the defence. 

Bangladesh has also ratified the ICCPR. Article 14 of the ICCPR, as interpreted by the Human Rights Committee in General Comment No. 32, requires equality of arms and an adversarial process in criminal trials: both sides must have a reasonable opportunity to present their case and challenge the evidence, and the defence must have access to the material on which the prosecution relies. General Comment No. 36, paragraph 41, makes clear that a death sentence imposed following an unfair trial amounts to an arbitrary deprivation of life. The United Nations’ Safeguards Guaranteeing Protection of the Rights of those Facing the Death Penalty further require that capital punishment, in countries which have not abolished the death penalty, may only be imposed “when the guilt of the person charged is based upon clear and convincing evidence leaving no room for an alternative explanation of the facts.”

Against this backdrop of domestic and international fair trial obligations, a 2024 amendment to the 1973 Act expanded the ICT’s framework for receiving and evaluating digital evidence. While section 19(1) allows the parties to rely on a wide range of prescribed class of digital materials — from CCTV, drone footage, and audio and video recordings to cell phone data and other records stored in hardware or software — section 19(1A) empowers the ICT to call for any evidence it considers necessary to determine the truth. Section 19(1B) requires the ICT to rule on admissibility and relevance by weighing probative value against any prejudicial effect on the fairness of the trial or on a fair evaluation of witness testimony, and section 19(1F) excludes evidence obtained in violation of internationally recognised human rights where the violation substantially undermines reliability or the integrity of the proceedings. Although these provisions weave substantive procedural safeguards into the digital evidence regime, seemingly aimed at ensuring both reliability and fairness in the adjudicative process, the force of those safeguards is qualified by provisions that grant the ICT considerable procedural flexibility. Section 19(5) states that the ICT “shall not be bound by technical rules of evidence” and may adopt “expeditious and non-technical procedures.” In a context where central evidence is digital and the death penalty is at stake, there appears tension between this invitation to informality and the obligation in section 6(2A) to ensure a fair trial.

Taken together with Bangladesh’s international human rights commitments, the fair trial guarantees contained in the 1973 Act establish that the ICT remains under a continuing duty to ensure a fair trial notwithstanding the restrictions imposed by Article 47A of Bangladesh’s Constitution. That duty assumes particular importance where the prosecution relies substantially on digital evidence and where a death sentence is potentially available, requiring scrutiny not only of the defence’s ability to review and challenge such evidence but also of the evidence’s provenance, handling, preservation, confidentiality, security, and overall integrity. Yet the effective discharge of that obligation appears to be hindered by the absence of a structured framework governing the collection, management, preservation, and evaluation of digital evidence, creating uncertainty as to how these safeguards are to be operationalised in practice.

A growing body of international standards and practice responds to precisely these concerns. The Berkeley Protocol on Digital Open Source Investigations (Berkeley Protocol), developed by the University of California at Berkeley and the Office of the United Nations High Commissioner for Human Rights, provides a principled framework for addressing precisely the evidentiary challenges associated with the collection, management, preservation, and evaluation of digital evidence. The Berkeley Protocol encourages reliance on digital open-source evidence in cases involving serious atrocities and international crimes, and sets out how to make that use responsible and fair. It sets out minimum standards for investigations of digital evidence, including requirements to document the chain of custody, preserve original files, record verification steps, and protect the dignity and safety of everyone captured in or affected by the footage. It is now influencing how international courts and investigative bodies treat open-source material, including at the International Criminal Court at The Hague, Netherlands, which has cautiously begun to integrate digital open-source evidence into its investigations of atrocity crimes. Therefore, it acts as a useful reference point for how to collect, verify, and preserve that material so that it stands up in court.

Similarly, the Leiden Guidelines on the Use of Digitally Derived Evidence in International Criminal Courts and Tribunals (Leiden Guidelines) identify broader evidentiary principles drawn from the practice of international criminal courts and tribunals, including in relation to videos, photographs, satellite imagery, intercepts, call data records, audio recordings, and other digitally-derived evidence. It, like the Berkley Protocol, emphasises transparency about methods and the need for judicial actors to understand the limitations of technical tools. Moreover, developed by Bellingcat and the Global Legal Action Network, the Methodology for Online Open Source Investigations (Bellingcat and GLAN Methodology) likewise offers a set of standard operating procedures complementing the Berkeley Protocol, aimed at translating open-source investigation techniques into legally usable practice by emphasising structured collection, preservation, verification, documentation, and disclosure. 

Although domestic courts in Bangladesh are under no positive obligation to follow the Berkeley Protocol, the Leiden Guidelines, or the Bellingcat and GLAN Methodology, they retain broad discretionary authority to refer to such soft law mechanisms, especially where hard law is inadequate or silent. Read together with the foundational principles of equality before the law and a fair trial, section 19 of the 1973 Act creates space, and arguably a duty, for the ICT to give concrete meaning to “probative value,” “prejudice,” “reliability” and “integrity” by drawing on international soft law. For instance, the Berkeley Protocol’s guidance on preserving originals, documenting chain of custody, recording verification steps, and minimising harm to witnesses and victims offers precisely the kind of detailed benchmarks that can operationalise section 19, without displacing the domestic framework. In that sense, looking to these frameworks is not an external imposition but a way of interpreting existing obligations in line with Bangladesh’s own fair trial commitments; and there is nothing stopping the tribunal from interpreting them through the lenses of these frameworks. The Supreme Court of Bangladesh has, in other contexts, acknowledged that international human rights standards and protocols, although not directly enforceable unless incorporated into national law, can serve as persuasive interpretive aids, especially where domestic law is insufficiently instructive (Bangladesh National Woman Lawyers’ Association vs Bangladesh 29 BLD 415; The Chief Prosecutor vs Abdul Quader Molla 8 LM (AD) 2020 375). Hence, the ICT may legitimately look to these non-binding international frameworks as persuasive interpretive aids in assessing the reliability, integrity, and probative value of digital evidence.

The question, then, is not whether digital evidence should be used in mass atrocity cases; it is clear that it must be. The question instead is how courts, defence counsels, and prosecutors should translate those investigative standards into evidentiary practice in a way that does not erode fair trial guarantees.

3. KEY CHALLENGES POSED BY DIGITAL EVIDENCE IN THE SHEIKH HASINA WAZED TRIAL

Before examining the three main challenges related to the ICT’s handling of digital evidence, a brief overview of the five charges of crimes against humanity against Hasina is warranted. The first charge alleges that, as the Prime Minister of Bangladesh and the President of the Bangladesh Awami League, she publicly branded quota reform protesters as “razakars” and by that incitement, and her subsequent failure to restrain security forces and party activists, triggered a widespread and systematic attack on student protesters that killed about 1,500 people and injured around 25,000. The second charge accuses her of conspiring, planning, and ordering law enforcement agencies to use lethal weapons, drones, and helicopters against unarmed demonstrators across the country. Charges three, four and five then link specific incidents to that broader campaign: the killing of student organiser Abu Sayed at Begum Rokeya University in Rangpur on 16 July 2024; the shooting dead of six protesters at Chankharpul near the Dhaka University on 5 August 2024; and the killing of five protesters in Ashulia that same day and the burning of their bodies, all allegedly carried out by police and other actors affiliated with Bangladesh Awami League under her incitement, conspiracy, planning, and orders, and her failure to exercise superior responsibility to prevent or punish these acts.

• AUTHENTICITY 
  

The first challenge is one of authenticity, concerning whether the digital material can be treated as reliable proof of the events it purports to depict. In the trial of Hasina before the ICT, digital evidence was used extensively. Journalists, police experts, and forensic specialists testified about videos and audio files. Wireless operators identified recordings of orders to use lethal force. The tribunal itself watched and listened.

For instance, journalist AKM Moinul Haque (Witness 5) testified as a prosecution witness that his live television broadcast from Begum Rokeya University on 16 July 2024 captured the incident of student Abu Sayeed being shot at close range in the chest and stomach, stumbling a few steps, and collapsing. He said the footage was aired live and later “seized” by the Police Bureau of Investigation. During cross-examination, the defence counsel asked only whether the videos might have been created using artificial intelligence (AI), which the witness simply denied; that was the extent of the exploration of manipulation for this crucial video. There is no indication in the record that the raw file’s metadata, upload history, or any other forensic indicators of alteration were examined in court. Other journalists, like photojournalist Muhid Hossain (Witness 30), submitted a pen drive with two videos documenting the shooting of his colleague Abu Turab in Sylhet on 19 July 2024. Here too, the videos were admitted as exhibits, yet there is little evidence of detailed in-court scrutiny of their provenance or integrity beyond the witness’s account that they were filmed while he was under fire. Measured against the benchmarks set by the Berkeley Protocol, the Leiden Guidelines, and the Bellingcat and GLAN Methodology, the presentation of evidence in this instance leaves a significant evidentiary gap: the court appears to have relied on the apparent visual content of the footage and the credibility of the submitting witness, without requiring a fuller account of how the files moved from the recording device to the courtroom, whether the files were original or copied, whether their integrity was preserved, and what verification steps were undertaken before admission.

The only systematic attempt to address AI manipulation concerns appears in the testimony of Sub-Inspector Md. Golam Iftekhar Alam (Witness 42), a photography and video expert at the Criminal Investigation Department (CID) of the Bangladesh Police. He received a pen drive with twelve videos and was asked to determine whether any of them were generated by AI. He testified that “no evidence was found that the videos received for examination were AI generated” and that he submitted his opinion to the investigating officer and a certified copy of this opinion was submitted to the court, which we have not reviewed. Nevertheless, the evidentiary foundation for that conclusion remained thin. Cross-examination revealed he had watched only one of the twelve videos and he did not know who had delivered or returned the pen drive. None of the videos he examined were actually played in court during his testimony. When asked if it was true that he had not correctly examined the received videos, he responded only, “I have seen them.” This falls short of the kind of transparent, replicable authentication process provided in the Berkeley Protocol and related standards. A bare assertion that material is not AI-generated, without disclosure of the tools used, the method applied, the limitations of that method, the files examined, and the chain through which those files reached the examiner, does not provide a sufficient basis for assessing authenticity in a capital atrocity trial.

Once viewed against these benchmarks, the issue is not simply that the defence failed to ask more probing questions. The more fundamental concern is that the prosecution and the tribunal did not appear to establish, on the record, a sufficiently transparent evidentiary basis for treating the videos as authentic and reliable.

These omissions reveal a twofold problem. First, the defence counsel did not use cross-examination to adequately probe gaps in the investigative and forensic processes, including provenance, metadata, chain of custody, AI-detection methods, and the limits of the expert’s conclusions. This sits uneasily with section 17 of the 1973 Act, which guarantees the accused the right to conduct their own defence or be assisted by counsel, to present evidence, and to cross-examine prosecution witnesses. Second, the shortcomings in examination-in-chief, cross-examination, and judicial evaluation point to a broader policy gap: the existing legal framework does not explicitly require the parties or the tribunal to structure the presentation and testing of the authenticity of digital evidence in a manner that gives practical effect to the fair trial obligation in section 6(2A) or reflects the level of rigour embodied in external standards such as the Berkeley Protocol, the Leiden Guidelines, or the Bellingcat and GLAN Methodology. The result is a disjunction between statutory fair-trial safeguards on paper and their practical implementation in a capital trial involving complex digital evidence.

One of the most consequential pieces of digital evidence in the trial of Hasina consists of audio files said to contain her own voice ordering lethal force and issuing threats. The evidence underpins the finding that she incited, ordered, and later admitted responsibility for crimes against humanity. At least two forensic specialists from CID gave key testimony in this instance. First, Inspector Rokunzaman (Witness 44) testified that a disputed audio file on a DVDR matched a sample of Hasina’s voice stored on a hard drive, and that he also matched another male voice to politician Fazle Noor Taposh. When pressed, he admitted: “I do not know whether there is a match between one person’s voice and another’s voice. I do not know whether people can mimic another person’s voice.” He provided no explanation of his methodology, software, error rates, or the possibility of false matches, yet the ICT appears to have accepted his conclusion. This falls short of the transparency and replicability expected under the international standards discussed above. A reliable authentication process would ordinarily require the expert to explain what material was examined, whether the files were originals or copies, how their integrity was preserved, what tools were used, how the comparison was conducted, what assumptions informed the analysis, and what limitations or error risks remained. 

Second, the testimony of Sub-Inspector Shahad Jubayer Lawrence (Witness 45) went further. He described using “international standard software” based on LR Ratio scores. In re-examination, he stated that in the CID lab an LR score of 1 or above means the disputed and sample voices are “considered to be one and the same.” He then reported scores of 2.7514 for the match between Hasina and one disputed voice, and 1025.9954 and 432.88127 for two further matches between her sample voice and other audio files, with similarly high scores for co-accused politician Hasanul Haq Inu.

On paper, this sounds reassuringly scientific. Yet cross-examination unearthed further problems. No higher authority appears to have reviewed or signed off on his report. He could not recall who, if anyone, had verified his work. There is no mention of calibration, error margins, or the risk that AI-generated voices might reach similar scores, even though he conceded that voices can be generated by AI. The ICT nevertheless relied heavily on these opinions, stating that the conversations “have been examined in the forensic laboratory of CID and the concerned experts found that the conversations are genuine and not generated by artificial intelligence.” 

The problem is not that these recordings are necessarily inauthentic. On the contrary, some of the key audio files have also been examined outside the proceedings by multiple media and forensic teams – including an investigative report by Tech Global Institution (TGI) in collaboration with The Daily Star, as well as other investigations by outlets such as Al Jazeera and the BBC – which independently concluded that the files were not fabricated. The concern here, instead, is threefold. 

First, the prosecution did not draw on this wider body of verified materials or their methodologies to substantiate their case. Second, the defence counsel did not interrogate and use to its advantage the margin of error inherent in these verification processes as a basis to dispute the prosecution case during cross-examination. Third, the ICT appears to have accepted the authenticity of the recordings at face value, by not sufficiently interrogating the forensic standard of authentication. For a piece of evidence that effectively anchors findings that a former Prime Minister ordered lethal force and bears superior responsibility, good practice, such as those espoused by the Berkeley Protocol and related standards, would point towards a more robust evidentiary record: multiple, independently generated reports, clear disclosure of methods and error margins, and treatment of the audio as one strand in a wider matrix of testimonial and documentary proof rather than as a near self-sufficient basis for causality. This minimal scrutiny sits uneasily with the statutory duty of the ICT, under sections 6(2A) and 19(1B) of the 1973 Act, to ensure a fair trial and to weigh the probative value of digital evidence against its potential prejudice to the accused. Instead, the ICT relied on a single, thinly explained analysis by CID and treated its conclusions as dispositive.

Despite these concerns, the verdict held in sweeping terms that “[t]he conversations recorded in DVD, CD and the videos of the occurrences have been examined by the CID experts who found that the said conversations and videos are genuine and not generated by artificial intelligence.” The ICT then added that the pen drives were played in court and that “thereupon it appeared to us that the conversations are real, not fake and fabricated.” Based on the verdict and publicly available testimonial evidence, the forensic review appears limited in what it disclosed on the record, and the underlying forensic reports referred to in the verdict and witness testimonies are not available in the public domain. It appears that a partial and opaque forensic review, combined with the judges’ own impressionistic listening, have been treated as sufficient to overcome reasonable doubts about the provenance, authenticity, reliability, and overall integrity of the digital evidence. The critique here is therefore directed at the way the forensic conclusions were presented and relied upon in the proceedings. In a trial where digital evidence was central to findings of guilt, and where the death penalty was imposed, fair trial guarantees require the technical basis for authenticity findings to be disclosed and explained sufficiently during the proceedings to allow meaningful challenge by the defence and informed evaluation by the tribunal.

• CHAIN OF CUSTODY
  

The second challenge is continuity of the evidentiary trail. Generally, traditional chain of custody assumes that evidence moves along a relatively clear line from crime scene to laboratory to courtroom. Digital material rarely behaves so neatly, which is why scrutinising the chain of custody in digital forensics is a different kind of inquiry than the traditional one. Social media videos may pass through multiple individuals, devices, and platforms before they reach a courtroom. Files can be downloaded, clipped, reuploaded, and compressed. Added to that, platform policies and content moderation decisions can alter what remains visible. Activists and journalists may edit footage for advocacy, cropping out details they consider non-essential but that might matter legally. The trial against Hasina before the ICT again offers a useful illustration. Journalists and wireless operators handed over pen drives and discs to the police. Forensic officers apparently received those storage devices via unnamed “carriers” and sent them back the same way. When questioned about who transported the devices, how they were stored, and what exactly was examined, several witnesses could not answer.

For instance, the tribunal heard from a wireless operator, Kamrul Hasan (Witness 50), who testified that between 11:00 and 11:30 on 17 July 2024, the Dhaka Metropolitan Police Commissioner Muhammad Mainul Hasan “instructed the use of force and instructed to shoot with Chinese rifles to suppress the student public movement.” His testimony was anchored in an audio file played in court, which he identified as containing the Commissioner Hasan’s voice and his own relay of the message. The pen drive was admitted as a material exhibit. Yet again, there is little detail about chain of custody, the origin of the recording, or any steps taken to rule out editing. Both the Berkeley Protocol and TGI’s methodology are explicit that investigators should preserve original files, keep detailed logs of every step and retain copies where platforms might remove content. As addressed above, section 19 of the 1973 Act likewise requires the tribunal to rule on admissibility by weighing probative value against any prejudice to a fair trial and to exclude evidence obtained in violation of internationally recognised human rights where this undermines reliability or the integrity of the proceedings. Bangladesh’s obligations under Article 14 of the ICCPR as interpreted by the Human Rights Committee also implies a duty to ensure that the provenance and handling of digital exhibits are sufficiently transparent to allow meaningful challenge. However, courts also need to develop more structured evidentiary practices aligned with the Berkeley Protocol and related international standards, grounded on clear accounts of the full life cycle of a digital exhibit, from the device that captured it, through platforms and intermediaries that hosted it, to the storage device that arrives in court. Without that, it becomes difficult to know which version of a video is actually being evaluated and whether it still reflects the scene it purports to show.

• PRESUMPTION OF INNOCENCE AND EQUALITY OF ARMS
  

The third challenge refers to the core criminal law guarantees of presumption of innocence, burden of proof, and equality of arms. Digital evidence can appear compelling on its face, particularly when it consists of graphic video footage, audio recordings, or forensic conclusions presented in technical language. For that reason, courts must ensure that such material does not circumvent the adversarial process or displace the prosecution’s burden of proof. Under both domestic and international human rights law, every defendant has the right to be presumed innocent, with the onus squarely on the prosecution to prove every element of the charges beyond reasonable doubt. Generally, the defence is under no obligation to present expert evidence at the outset; its primary function is to test, challenge, and contextualise the evidence relied upon by the prosecution.

During the trial, the defence counsel for Hasina tried at various points to question the technical basis of the forensic opinions and even asked for more time to “gather knowledge” before cross-examining one of the experts. The prosecutor objected that this was unnecessary. The tribunal encouraged the defence to proceed immediately and later held that a single national forensic laboratory examination was enough, dismissing the defence argument that independent or additional testing should have been ordered. Throughout, Hasina was tried in absentia and represented only by a state-appointed lawyer whom she apparently had not chosen and who appears to have had no independent digital expertise at their disposal. As discussed above, section 17 of the 1973 Act recognises the accused’s right to a fair trial, to be informed of the charges and to be represented, to present evidence and to cross-examine witnesses, but in practice these safeguards appear to have been applied in a highly formalistic way that did little to offset the structural disadvantages faced by the defence in a trial with complex digital evidence.

Defence had no independent expert witness testifying on their behalf. Rather, their only question to one crucial witness about AI manipulation was a single line: “Is it possible that the videos you submitted were created using artificial intelligence.” The answer was a simple denial, and no one probed what that would actually mean in practice. There was no exploration of what forms AI manipulation might take, how such manipulation could be detected, or whether any steps had in fact been taken to rule it out. A proper cross-examination or judicial assessment of possible AI manipulation would not require the defence to prove that the material was fabricated. It would, however, require targeted questions capable of testing the basis for the prosecution’s claim of authenticity. At minimum, this would include questions about whether the original or native file was examined; whether metadata, hash values, compression patterns, timestamps, and upload history were reviewed; what software or forensic tools were used; whether those tools are capable of detecting synthetic or AI-altered content; what error rates or limitations apply; whether the footage was compared with other verified footage, witness accounts, geolocation, chronolocation, or environmental cues; and whether the expert could explain how manipulation was ruled out rather than merely asserting that it was. For a capital trial, this raises serious concerns not only about resources, but also about whether state-appointed counsel discharged their professional duty of diligence and zealous representation, as required under the Bangladesh Bar Council’s Canons of Professional Conduct and Etiquette. If obvious gaps in the prosecution’s forensic case are left untested, the presumption of innocence risks becoming a legal fiction rather than a lived trial principle. 

Under the Berkeley Protocol, investigators are expected to maintain meticulous records of how digital items were acquired, processed, and preserved, so that courts and defence teams can replicate and test their findings. Likewise, the Leiden Guidelines emphasise transparency about methods and the need for judicial actors to understand the limitations of technical tools. So too does the Bellingcat and GLAN Methodology, which stress multi-source verification, detailed documentation, and disclosure of error margins and limitations. By contrast, the process in this case appears to have relied primarily on expert assurances and judicial acceptance, without the level of methodological transparency, documentation, and adversarial testing contemplated by these frameworks.

This goes to the heart of equality of arms, articulated in Article 14 of the ICCPR. As noted above, General Comment No. 32 emphasises that a fair trial requires equality of arms and an adversarial process, whereby the parties are placed on an equal footing in presenting their cases, testing the evidence, and, in the case of the defence, obtaining access to the prosecution’s evidentiary materials. Notably, the form of the evidence does not change those obligations; and in a capital case, these guarantees are inseparable from the right to life. The United Nations’ Safeguards Guaranteeing Protection of the Rights of those Facing the Death Penalty state that capital punishment may be imposed only where guilt is based on clear and convincing evidence leaving no room for an alternative explanation of the facts. Additionally, the Human Rights Committee’s General Comment No. 36 clarifies that imposing the death penalty following proceedings that do not comply with the fair trial guarantees enshrined in Article 14 of the ICCPR amounts to an arbitrary deprivation of life under Article 6 of the ICCPR. Indeed, the persuasive power and technical complexity of audio-visual material mean that courts must be even more attentive to ensuring that digital exhibits can be meaningfully tested. If digital material is going to play a central role in proving guilt, then the right to challenge that material needs to be taken as seriously as the right to confront a live witness. Viewed against these standards, the combination of an in-absentia trial, reliance on a single state forensic laboratory, limited cross-examination, and the absence of independent defence expertise may raise questions about whether the threshold of proof beyond reasonable doubt was fully satisfied in establishing the defendant’s guilt.

4. IMPLICATIONS 

Despite the challenges presented, these strands of digital evidence were treated decisively. The ICT concluded that Hasina ordered the use of drones, helicopters, and lethal weapons, and that law enforcers and ruling party activists killed around 1,500 protesters and injured about 25,000. It held that she bore superior command responsibility and sentenced her to death. In doing so, it treated what appears to be contested digital evidence as sufficiently clear and convincing to justify the most irreversible of punishments, despite the unresolved questions about authenticity, provenance, and equality of arms.

The explosion of digital evidence from the July Revolution has been vital for documenting killings that the Bangladesh Awami League tried to deny or downplay. It has given bereaved families something to hold on to. It has undercut attempts to erase what happened from public memory. At the same time, the trial of Hasina before the ICT illustrates a different concern: the way the tribunal engaged with, tested, and relied on the digital evidence within the proceedings. Where digital footage is treated as self-authenticating, or where its provenance, forensic basis, and limitations are not adequately explained on the record, fair trial guarantees can be strained even when the underlying material is powerful and highly probative. Viral clips, dramatic audio files, and other forms of digital evidence have played a crucial role in documenting abuses, preserving collective memory, and supporting accountability efforts, often at significant personal risk to those who recorded or shared them. At the same time, their evidentiary value in court depends on careful authentication, contextualisation, and adversarial testing. When judges and lawyers lack the technical tools to explain that distinction, there is a real risk that the emotional force of watching a shooting or hearing a lethal command may be relied on in place of properly tested and contested proof beyond reasonable doubt. 

Precisely because digital evidence can be so compelling to victims and the public, courts in atrocity trials have an even greater responsibility to insist on rigorous verification, clear explanations of what the technology can and cannot do, and meaningful opportunities for challenge before treating it as probative of guilt.

This is not to suggest that Bangladeshi investigators or prosecutors lack digital expertise. Reports indicate that the ICT prosecution team has drawn on digital forensic capacity, including the appointment of a cybersecurity and digital forensic investigator. Additionally, broader capacity-building initiatives for Bangladesh law enforcement agencies, including in cooperation with the United Nations Office on Drugs and Crime, have also appears to have addressed cybercrime and forensic skills. As such, the issue is not the absence of local expertise, but whether that expertise is embedded in a transparent evidentiary framework that requires preservation, disclosure, independent testing, and fair-trial safeguards in court.

The challenges highlighted in this policy brief are sharpened by the fact that sophisticated digital expertise is still relatively rare, especially in under-resourced jurisdictions in Global Majority contexts. Authenticating, analysing, and evaluating digital evidence often requires specialised software, hardware, and training that most domestic laboratories cannot sustain on their own. In other atrocity contexts, that gap has sometimes been bridged through structured collaboration with external digital investigation units in the United Nations mechanisms, international human rights organisations, or academic labs. 

While acknowledging that many victims’ families wish to see capital punishment in cases involving grave harm and mass violence, it substantially narrows the scope for such external collaboration. Many of the organisations with the strongest digital verification capacities are bound by mandates or ethical policies that preclude assisting proceedings that may result in an execution. The case against Hasina illustrates this structural trade off: a statute that retains the death penalty can impede the tribunal’s access to independent digital expertise and independent technical assistance that would otherwise strengthen both the reliability of the evidence and the perceived fairness of the process. 

Set against the ICT’s institutional history, the case against Hasina also exemplifies the continuity of retributivist and exceptionalist legacy of the tribunal rather than a genuine attempt to establish accountability for atrocity crimes. The ICT was re-established by Hasina herself to prosecute alleged collaborators with the Pakistan Army during Bangladesh’s 1971 Liberation War. These trials were similarly criticised for failing to meet international standards on due process and for its persistent use of the death penalty. Although the interim government has amended the 1973 Act to bring it more in line with international criminal law and has updated its procedures to accommodate digital evidence, it has not undergone a significant structural or ideological shift from the period when it adjudicated cases under the Bangladesh Awami League government. 

In a written statement, Hasina swiftly condemned the verdict, arguing that the prosecutors failed to produce persuasive evidence to show that she had ordered the use of lethal force against protesters, or afford her a “fair chance to defend [herself] in court, nor even to have lawyers of [her] own choice represent [her] in absentia.” She further alleged that the “[t]ranscripts and audio files cited as evidence are fragmentary and have been taken out of context. The fact is that operational control rested with security forces on the ground, acting under established legal protocols.”      The irony is stark: a tribunal once weaponised by her regime to send political opponents to death row has now been turned against its own architect, using the same legal machinery she helped entrench.

Precisely because the ICT operates within a statutory framework that at least formally recognises fair trial guarantees, these procedural deficiencies have direct legal implications by providing grounds for challenging the verdict. If political dynamics shift in the future, the decision could be challenged. One need only consider the ICT’s own history, which already contains examples of convictions or sentences being modified or overturned by the Appellate Division of the Supreme Court of Bangladesh on grounds of evidentiary weakness, procedural irregularities, or inadequate reasoning. Against that backdrop, a capital judgment grounded in thinly tested digital evidence and an in-absentia process may prove far less durable in law than it currently appears in politics.

It appears that trials in absentia, capital sentencing powers, and a highly deferential approach to prosecution evidence remain intact in the functioning of the ICT. However, that continuity is deeply troubling for any genuine truth, justice and reconciliation project and sits uneasily with the interim government’s stated commitment to institutional reform. At the same time, the Hasina case sets a precedent that other Global South jurisdictions may look to when prosecuting serious crimes: one in which digital evidence is admitted and relied upon, but the procedures for testing that evidence, and the safeguards surrounding its use, remain underdeveloped. If that approach becomes the template, the digital turn may also entrench older habits of instrumentalising courts in moments of political transition without fully protecting fair trial rights. 

In contrast, the Monsoon Protest Archives, co-developed by TGI, International Truth and Justice Project, and Netra News, demonstrate an alternative approach and methodology for the collection and handling of user-generated digital evidence in atrocity documentation, based on a digital forensics investigation manual designed in alignment with internationally recognised standards. Having collected, analysed, verified, and preserved thousands of images and videos from publicly accessible sources and eyewitnesses during the 2024 July Mass Uprising, TGI and its partners endeavoured to create a database following a documented and unbroken chain of custody from collection to secure storage, while applying advanced verification methods, such as geolocation via visual landmarks and cues, chronolocation through shadow and environmental analysis, metadata extraction, and expert consultations in weapons, law, and open-source intelligence, to confirm authenticity and rule out manipulation or artificial generation, where appropriate. Such methodologically rigorous standards strengthen evidence integrity, offering a model that upholds fair trial principles and enhances credibility in accountability efforts. They also show that more rigorous digital evidentiary practices were both conceptually and practically available in Bangladesh at the time of the Hasina trial.

5. RECOMMENDATIONS

Against this backdrop, several recommendations for reforms emerge from the trial of Hasina that are relevant not only for Bangladesh but may also offer useful reference points for any jurisdiction, especially in the Global South, seeking to use digital evidence in atrocity prosecutions without sacrificing fair trial guarantees.

TO THE GOVERNMENT OF BANGLADESH 

• Constitutional Reform: Amend or repeal Article 47(3) and 47A of Bangladesh’s Constitution to dismantle the selective rights regime that carves atrocity suspects out of key fair trial protections, restoring full access to fundamental rights in such proceedings, including Articles 31 (right to protection of law) and 35 (protection in respect of trial and punishment).
• Death Penalty Moratorium: Declare an immediate moratorium on the death penalty, pending full abolition. Such a measure would bring Bangladesh into closer alignment with the increasing number of states that have abolished the death penalty in law or practice. It is a practical necessity to unlock cooperation with international digital forensic bodies and United Nations mechanisms that are legally barred from assisting in capital cases.
• Reforms to the 1973 Act: 
• Fair Trial Supremacy: Insert a “supremacy” clause clarifying that in cases of conflict between the admissibility of digital evidence and the rights of the accused, the fair trial obligations in section 6(2A) and compatibility with international human rights law in section 19(1F) shall prevail.
• Tighten Evidentiary Rules: Amend section 19(5) to define strictly what “non-technical procedures” entail and clarify that flexibility cannot override the minimum standards required for verifying the authenticity of digital evidence.
• Resource Allocation: Amend section 17 to require state funding for independent digital forensic assistance for the defence, including access to qualified experts who can review the collection, preservation, authentication, and presentation of digital material, examine the underlying files and metadata, and provide reports that can be tested through cross-examination. This would help ensure true equality of arms, particularly for indigent or absent defendants and would make fair trial guarantees operational in cases involving complex digital evidence.
  

TO THE JUDICIARY OF BANGLADESH 

• Adoption of Practice Directions: Issue a formal practice direction on digital evidence that explicitly adopts the Berkeley Protocol or comparable frameworks as the standard for admissibility and handling digital evidence. This guidance should require, amongst other things, presentation of the original raw files (not just re-saved copies), full documentation of the chain of custody (who held the device, how files were transferred), and disclosure of verification methodology (software used, error rates).
• Higher Standard of Proof for “AI” Claims: Reject bare assertions that evidence is “not AI-generated” based solely on visual inspection or opaque software scores. Instead, require a provenance-aware assessment, drawing on existing resources, for instance, those developed by the Coalition for Content Provenance and Authenticity (C2PA) and WITNESS. The ICT should require forensic reports to explain, in concrete terms, how manipulation was ruled out, including what files were examined, whether original or copied files were used, what metadata and transfer history were reviewed, what technical tools or methods were applied, what limitations those methods have, and whether the material was checked against other verified footage or contextual evidence. Both the report and the forensic experts should be made available to defence counsel for meaningful cross-examination. 
• Active Judicial Inquiry: In cases of absentia trials or state-appointed defence, judges should adopt a more pro-active inquisitorial role, actively questioning the provenance of digital evidence rather than accepting the prosecution’s submission at face value. 
  

TO THE PROSECUTION AND INVESTIGATION AGENCIES 

• Transparent Forensic Methodology: Move beyond “black box” reporting (e.g., LR Ratio scores without context) and adopt a more robust forensic approach to digital evidence. Forensic reports submitted to the court must explain the software used and its industry accreditation, the known error margins and limitations of the tool, and whether the software has been tested against current generative AI capabilities.
• Chain of Custody Integrity: Build on the ICT’s existing use of hash value mechanisms by requiring that hash-values be recorded immediately upon seizure of digital devices and re-checked at every later stage of handling, including access, copying, transfer, and storage, with contemporaneous logs sufficient to verify the full custody trail and detect alteration or substitution. 
• Separation of Functions: Establish a firewall between the investigative team (police and other law enforcement agencies) and the forensic analysts to minimise confirmation bias. Where independent domestic laboratories are not yet available, the ICT should at least require that forensic analysis be documented in a way that allows scrutiny by the court and defence, and should expressly permit the defence to seek independent testing through accredited external laboratories or qualified experts, subject to any lawful security or procedural constraint
  

TO THE DEFENCE COUNSEL (BOTH STATE-APPOINTED AND PRIVATE)

• Digital Literacy and Zeal: Uphold the Bangladesh Bar Council’s Canons of Professional Conduct and Etiquette by adequately testing digital evidence. This includes requesting the native or original file where available,  insisting on disclosure of metadata and transfer history, challenging any gaps in chain of custody or forensic method, and declining to accept authenticity without a defensible basis in the record.
• Demand for Experts: File motions requesting court-appointed independent digital experts where the case turns on technical evidence, particularly in capital or in-absentia proceedings. Where a court-appointed expert is not feasible, counsel should seek permission for independent testing through accredited external laboratories or qualified specialists, so that the defence has a realistic opportunity to test authenticity, provenance, and reliability.

TO THE CIVIL SOCIETY 

• Adoption of Archival Standards: Follow international best practices and standards on analysis of digital evidence when documenting human rights violations, prioritising preserving metadata and hashing files immediately upon collection to create a “shadow” chain of custody that can later corroborate or contradict official state evidence.
• Legal Training: Conduct workshops for Bangladeshi lawyers (both prosecution and defence) and judges on the technical basics of digital forensics, how to read metadata, how to spot deepfakes, and how to interrogate a digital expert.
• Shadow Reporting: Actively monitor ICT trials specifically for digital due process violations, and publish shadow reports analysing the technical validity of evidence accepted by the court to maintain public scrutiny on the process.
• Appointment of Independent Digital Experts as Amici Curiae: Advocate for the formal appointment of independent digital forensics experts as amici curiae to assist the ICT in assessing the authenticity, provenance, and integrity of digital evidence, particularly in cases involving complex technical questions. Such experts should be institutionally independent from the prosecution and defence, and mandated to provide neutral, court-facing expertise to enhance judicial understanding and safeguard fair trial guarantees.
  

TO THE INTERNATIONAL COMMUNITY

• Knowledge Transfer: Facilitate “South-South” partnerships where experts from other Global South jurisdictions who have handled digital evidence in atrocity trials can mentor Bangladeshi practitioners, ensuring standards are adapted to the local context but remain rigorous.
• Conditional Technical Assistance: Build on existing domestic expertise and any prior capacity-building initiatives by offering targeted technical assistance, peer exchange, and training on the collection, preservation, verification, disclosure, and courtroom testing of digital evidence, provided that such support is tied to compliance with international human rights and fair trial standards.

6. CONCLUSION

This policy brief has attempted to show that the ICT’s handling of digital evidence, when measured against emerging international standards and best practice, exposes three structural vulnerabilities of the digital turn in atrocity trials: opaque forensic methods, fragile chains of custody, and entrenched inequality of arms between prosecution and defence. These challenges are not unique to Bangladesh. Rather, they are increasingly relevant to atrocity investigations and prosecutions in conflict-affected regions around the world, including Palestine, Iran, Syria, Ukraine, and elsewhere, where digital evidence is playing an ever more prominent role in documenting and adjudicating mass atrocities. The main task is therefore to ensure that digital evidence strengthens accountability rather than weakening fair trial standards. As evidence before atrocity trials becomes increasingly digitalised, it is crucial that we remain honest about that tension. If these kinds of reforms are taken seriously, the digital turn in atrocity prosecutions can help close long-standing evidentiary gaps without reopening old wounds around selective justice and unfair trials. The choice is not between using digital evidence and protecting fair trial rights, but between doing both badly and doing both well.