Deepfakes, Democracy, and the Cost of Digital Deception in Bangladesh
In late February 2026, a video began quietly circulating on Facebook where Prime Minister Tarique Rahman stood at what appeared to be an official gathering, handing bundles of cash to uniformed army members.
In his own synthesised voice, he addressed the camera, saying, “Countrymen, I am giving Family Cards and money to the army to distribute to you. Those who wish to receive them, please follow, share, and comment.”
Within days, 52 such videos had spread across at least 31 Facebook pages. One featured Tarique’s daughter Zaima, claiming she was travelling village to village with Tk 20,000 in hand.
At the time of investigation, one fake page bearing Zaima’s name had accumulated 57,000 followers. Another had 32,000 — all within a single week.
None of it was real. Every video was fabricated using artificial intelligence or AI. Every promise was a trap. Every follower was a mark.
The victims were not foreign governments or geopolitical rivals. They were ordinary Bangladeshi farmers, garment workers, and shopkeepers who saw the face of their PM, heard his voice, and believed. They shared their bKash numbers, their addresses, and their names. They waited for money that would never come.
What Is a Deepfake — And Why Does It Work So Well Here?
A deepfake is an AI-generated video or image in which a real person’s face, voice, or both have been seamlessly replaced with fabricated content. The technology, which first emerged publicly in 2017, has become dangerously accessible. Tools that once required sophisticated computer hardware and months of training data can now be rented online for a little money per month. A convincing video of a public figure can be produced in hours, sometimes minutes.
In high-literacy, high-trust media environments, deepfakes can still do damage. But fact-checkers, platform moderators, and a sceptical public provide a partial buffer. Bangladesh does not yet have that buffer.
Management and Resources Development Initiative found that news literacy among Bangladeshi citizens runs as low as 76 percent. It means that a significant portion of the population lacks the basic critical tools to question the source, context, or authenticity of what they see online. Facebook, the primary digital public square for most Bangladeshis, hosts an audience where a verified-looking video of a known political leader can spread to millions before a single fact-checker has even seen it. By the time any correction arrives, the electoral, financial, or reputational damages are already done.
The 2024 Election: A Preview of the Threat
In the 2024 general election, AI-generated content quietly entered the electoral arena.
An investigation by the Financial Times, later confirmed by Bangladeshi fact-checkers at BOOM and Dismislab, found that the then pro-government media outlets and influencers had produced AI-generated disinformation using cheap, commercial tools. In one case they used a platform called HeyGen and presented a studio segment accusing US diplomats of interfering in Bangladesh’s elections and causing political violence. The segment looked, to an untrained eye, like a genuine broadcast.
Other examples include anti-opposition deepfake videos posted on Meta’s Facebook, including one that falsely purports to be of an exiled BNP leader at that time, suggesting the party “keep quiet” about Gaza to not displease the US. The Tech Global Institute, a think tank, and media nonprofit Witness both concluded the fake video was likely AI-generated.
Selling Lies: How Deepfakes Exploit Ordinary People
For many Bangladeshis, the political elite is a distant abstraction. The local Member of Parliament, the Prime Minister, and the party leader exist primarily as images on screens, on banners, and in Facebook videos. When those images speak, they carry authority that is difficult to question.
This is what the deepfake exploits.
The Family Card fraud of early 2026 was not a sophisticated operation. The videos were sometimes clumsy up close, like lips slightly out of sync, skin texture slightly too smooth, and background details inconsistent. But to someone scrolling on a budget smartphone with a small screen and a weak data connection, none of that was visible. What was visible was the face of their PM, promising help.
When deepfake videos urge users to share specific content, people with limited digital literacy often become involved and help spread it further. This allows the content creators to generate revenue. At the same time, personal information like phone numbers, bKash numbers, and addresses ends up in the hands of scammers.
The person who shared the video was not complicit. They were deceived. And in sharing it, they became an unwilling amplifier, carrying the fraud further into their own social circle, to relatives and friends who trusted them.
This is the compounding geometry of deepfake deception. One fabricated video does not reach one million people because it was seen by one million sceptics. It reaches one million people because each believer told five others they could trust. The deepfake’s most powerful distribution mechanism is human trust.
A Society on Uncertain Ground: The Broader Social Impact
1. The Erosion of Evidential Truth
Perhaps the most insidious long-term consequence of deepfakes is not any single fabricated video, it is what they do to the concept of evidence itself.
When a genuine video of wrongdoing emerges — an official taking a bribe, a public figure acting violently, a politician making a damaging admission — the deepfake creates an instant alibi. Any powerful figure can now respond to authentic footage by claiming it is fake. Their allies amplify the denial. Fact-checkers may eventually confirm the video’s authenticity, but by then, the public is caught in a fog of competing claims.
This is what researchers at the Tech Global Institute call the “Liar’s Dividend” — the indirect benefit that deepfakes deliver not by fabricating specific content, but by poisoning the well of trust in all digital content. In a country where accountability journalism is already under pressure and where courts and institutions can be slow to verify technical evidence, this dividend is particularly dangerous.
2. Political Paralysis Through Confusion
When a population cannot reliably determine what is real, political engagement becomes much harder. Voters in Bangladesh already contend with limited access to independent media, high levels of disinformation on social platforms, and, in many cases, direct political intimidation. Deepfakes add a further layer of disorientation.
If a candidate’s alleged statement can always be disputed as fabricated — even when it is genuine — political accountability becomes nearly impossible to establish. The powerful benefit from this confusion; the ordinary citizen does not.
3. Chilling Effects on Political Participation
Deepfakes do not only affect election outcomes. They shape who is willing to participate in public life at all. When activists, journalists, and ordinary citizens see what can be done to a public figure’s reputation with a few hours of AI work and a Facebook account, many choose silence. The threat does not need to be carried out. Its mere existence is sufficient to deter.
This chilling effect is already documented in Bangladesh’s digital environment. Researchers have tracked how online political discourse, particularly dissent and criticism, has contracted under the combined pressure of cyber laws and the threat of digital character assassination. Deepfakes accelerate this process.
4. Financial Fraud at Scale
The Family Card scam is one documented case. It is not isolated. As deepfake tools become cheaper and more accessible, the barrier to financial fraud using fabricated political content is essentially zero. Bangladesh’s growing mobile financial infrastructure – bKash, Nagad, and similar platforms – means that millions of people are one trusted video away from handing over real money for a fake promise.
The demographic most at risk is precisely the population that political leaders most want to reach: rural, semi-urban, lower-income voters who follow politicians on Facebook, who watch political videos for news, and who have no other reliable verification mechanism available to them.
What Is Being Done — And What Is Not Enough
Bangladesh is not standing still. In May 2025, the interim government enacted the Cyber Security Ordinance 2025 — replacing a deeply criticised predecessor law that had been used primarily to suppress opposition. The new ordinance takes meaningful steps: it criminalises online sexual harassment, revenge pornography, blackmail, and – crucially – cybercrimes committed using artificial intelligence. Its advocates describe it as the first such law in South Asia.
These are genuine advances. But legal frameworks, however well-designed, face serious structural challenges in Bangladesh’s current environment.
Evidence handling is a persistent problem. Digital evidence in Bangladeshi courts requires a mandatory certificate under Section 65B(4) of the Evidence Act. Without it, the evidence is inadmissible. Deepfake content can be altered or deleted in hours. Cases frequently collapse not because the crime did not occur but because digital evidence was not preserved in time or because judicial officers lack the technical literacy to evaluate it.
Platform accountability remains weak. Meta, which operates Facebook – the primary vector for deepfake spread in Bangladesh – has responded to high-profile cases only after significant media pressure and rarely proactively. The company’s content moderation in Bangla is inadequate relative to the volume of content being produced and shared. Smaller markets, as researchers have consistently noted, receive less attention than English-language ones.
What Can Actually Be Done: A Path Forward
The deepfake crisis in Bangladesh is not primarily a technology problem. It is a literacy problem, an institutional problem, a political problem, and a media problem — and it requires responses at each of those levels.
Digital and Media Literacy — Urgently, at Scale
The most immediate and cost-effective intervention is education. A population that can identify the hallmarks of manipulated content — unnatural lip-sync, skin-texture anomalies, and inconsistencies in background — is a population that can slow the spread of deepfakes before they take root.
This means integrating media literacy into school curricula beginning at the primary level. It means running sustained public awareness campaigns — not one-off announcements but ongoing, multi-platform efforts delivered in Bangla across television, radio, Facebook, and community gatherings. Bangladesh has a significant youth population with high levels of social media engagement. That same network that currently accelerates deepfake spread can, with the right literacy tools, become a network of informal fact-checkers.
Civil society organisations and independent media houses have a critical role to play here. International models — including the EU’s media literacy initiatives and fact-checking networks built around partnerships with academic institutions — offer templates that can be adapted.
Institutional and Legal Reform
The law must be implementable, not merely symbolic. This requires investment in two areas: technical capacity within the judiciary and law enforcement, and procedural reform to ensure digital evidence can be gathered, preserved, and presented effectively.
Judges and prosecutors need dedicated training in digital forensics. Police cyber units need both funding and accountability — the current pattern where most cyber complaints are filed and forgotten is unsustainable as deepfake incidents multiply. A dedicated fast-track process for deepfake-related cases, with clear timelines and specialist adjudicators, would significantly improve deterrence.
On the legislative side, the Cybersecurity Ordinance 2025 requires implementing regulations that address deepfakes specifically — including a clear legal definition of synthetic media, provisions for rapid content removal with judicial oversight, and whistleblower protections for those who report deepfake content professionally.
Platform Accountability and Transparency
Tech platforms — primarily Meta — must be held to higher standards of accountability in the Bangladeshi market. This means mandatory reporting of deepfake-related content removal in Bangla-language content, faster response times to takedown requests from verified civil society organisations, and investment in Bangla-language AI detection tools.
Bangladesh can learn from the approaches taken by the European Union under its Digital Services Act, which requires large platforms to conduct and publish risk assessments for electoral integrity – and to cooperate with national regulators in addressing identified risks. Advocating for similar requirements through bilateral and multilateral channels is a legitimate and productive policy goal.
Independent Fact-Checking Infrastructure
Bangladesh has a small but credible ecosystem of fact-checking organisations — Dismislab, BOOM Bangladesh, Digitally Right — that have demonstrated the ability to identify and debunk deepfakes under significant resource constraints. This ecosystem needs sustainable funding, legal protection from retaliatory suits, and institutional recognition.
A national consortium – bringing together independent fact-checkers, university media departments, civil society organisations, and platform representatives – could serve as an early-warning system for deepfake content during elections and political moments. Election monitoring frameworks in future cycles should explicitly include AI and synthetic media in their scope.
Political Will and Cross-Party Norms
Ultimately, deepfakes become most dangerous when political actors use them as weapons — or when they use the possibility of deepfakes to discredit legitimate accountability journalism. What is needed, alongside all technical and legal measures, is a political norm: a cross-party commitment that no political entity in Bangladesh will commission, produce, or amplify deepfake content.
Such norms are difficult to enforce legally. But they can be built through public naming and shaming, through civil society pressure, and through a media culture that treats the use of AI disinformation as a serious breach of political conduct. The 2024 experience showed that Bangladesh’s political environment is not ready for that norm yet. Building it is a long-term project. But it must begin.
The Reckoning Ahead
The Tarique Rahman Family Card videos were, in a sense, unsophisticated. The scammers were motivated by money, not politics. The fabrications were relatively detectable. And yet they reached tens of thousands of people within days, harvested real personal data, and exploited the genuine trust that ordinary Bangladeshis place in their political leaders.
What happens when the next election cycle begins — with real competition, real stakes, and real political actors willing to commission more sophisticated content? What happens when a deepfake of a leader making an incendiary statement circulates 48 hours before voting, too late for any fact-checker to reach the rural voter who has already seen it?
Bangladesh is not uniquely vulnerable. Similar dynamics are playing out across South Asia, across Africa, across every democracy where digital access has outpaced digital literacy. But Bangladesh faces a particular confluence of pressures: a highly polarised political environment, a low-literacy media landscape, an enormous Facebook-dependent public sphere, and a legal system that is only beginning to develop the tools to respond.
The technology is not going to slow down. The tools will get cheaper. The fakes will get better. The question is not whether Bangladesh will face a deepfake-driven political crisis. The question is whether, when it comes, the country will have built enough of the institutional, civic, and legal infrastructure to survive it with its democracy intact.
The window to build that infrastructure is open. It will not stay open forever.
