State surveillance and policing of online expression have a long history in the country
Originally published on Global Voices
Photo by cottonbro studio. Used under a Pexels License.
This post is part of Global Voices’ October 2026 Spotlight series, “Big Tech and Authoritarianism.” In this Spotlight, we explore how Big Tech facilitates and hinders authoritarian governments and practices, and how these choices influence the Global South. You can support this coverage by donating here.
Bangladesh is revising its cyber law to tackle AI-generated images, and the draft has triggered a furious debate over how much the state should police speech online. The ICT Division published the draft Cyber Protection (Amendment) Act for public comment in mid-September. It is now with a five-member cabinet committee led by Posts, Telecommunications and ICT Minister Faqir Mahbub Anam. In a discussion reported on October 19, media figures, legal analysts, and civil society representatives called for the most contested clauses to be removed.
The push follows a campaign season crowded with synthetic media. Global Voices previously reported on a study that identified 72 AI-manipulated items that gained traction between December 2025 and the February 12, 2026 election. Synthetic images of political figures were the most common tactic, alongside doctored news-channel graphics. The Bangladesh Nationalist Party (BNP), which went on to win, was the most frequent target, with 47 cases.
State surveillance and policing of online expression have a long history. Section 57 of the Information and Communication Technology (ICT) Act of 2006 punished online content seen as defamatory, hurtful to religious sentiment, or damaging to the state’s image. A 2013 amendment toughened it, with sentences of up to 14 years. Journalists and ordinary social media users were among those charged under it. After years of protest, the government promised to scrap Section 57. Yet when the Digital Security Act (DSA) was signed into law in October 2018, the same offenses reappeared in sections 25, 28, 29 and 31. By 2023, the law minister had told parliament that 7,001 cases had been filed under it. The writer Mushtaq Ahmed died in custody in 2021 after ten months in jail under the DSA, which led to protests at home and abroad.
The Awami League government replaced the DSA with the Cyber Security Act in 2023, but Amnesty International said it repackaged almost all of the DSA’s repressive features. After the July 2024 uprising, the interim government scrapped nine sections through a new Cyber Security Ordinance in 2025. Rights groups say hundreds of cases from earlier years are still pending.
Fakes in the election and beyond
Women in politics faced a particular kind of attack. A Daily Star analysis of AI content before the vote found that pro-Awami League pages targeted female leaders of the National Citizen Party (NCP) with fabricated images that hinted at moral and sexual impropriety. After the election, a Prothom Alo fact-check found AI-generated images and invented statements aimed at nominees for reserved women's seats. Tanvir Mahatab Abir of Rumor Scanner told the paper that visible women, especially in politics, are frequent targets of disinformation.
Not all of it is political. Data from the Daily Star shows that Facebook pages used deepfake videos of Prime Minister Tarique Rahman and his family, including his daughter Zaima Rahman, to promote fake “Family Card” offers in exchange for likes and follows. Among 52 posts analyzed, 13 featured Zaima. One page administrator told the fact-checking outlet Dismislab that the aim was to gain followers quickly and sell the pages. In such cases, for many creators of synthetic content, monetization metrics on social platforms often serve as a primary driver alongside political motives.
A case that left no remedy
For ordinary women, the damage can last much longer. Farzana Sithi, a student activist known for her role in the historic student-led uprising in 2024, was targeted by a fake Facebook profile that Dismislab investigated in March 2025. Of 166 videos and reels posted between December 13, 2024, and March 17, 2025, 133 were AI-generated deepfakes. They placed her face on videos of two Indian content creators. The five most-viewed clips drew more than 4.1 million views, and the account was monetized through Facebook's Stars feature.
“These videos are an attempt to tarnish my character,” Sithi told Dismislab, adding that she was bullied and shamed as a result. She reported the profile to Facebook several times. IT experts she approached privately asked for USD 100 to USD 150 to remove the fake accounts and videos, which she said she could not afford. At the time of Dismislab’s report, the profile was still active. Comments under the reels included rape threats.
What the law covers, and what the draft would change
Legal experts point out that the existing legislative framework has struggled to keep pace with rapid advancements in synthetic media. The Cyber Protection Act was enacted on 10 April 2026 and amended on 30 June. Its Section 25 already covers blackmail, sexual harassment, revenge porn, child sexual abuse material, sextortion and the distribution of AI-generated or manipulated content.
According to the minister, these offenses carry up to two years in prison, or up to five if the victim is a woman or under 18. The law targets AI mainly when it is used to commit existing crimes. It does not set out a dedicated framework for synthetic media or political deepfakes.
The draft would change that. It adds defamation, humiliation and bullying to Section 25 and names AI-generated content explicitly. A prior investigation by Prothom Alo revealed that the maximum penalty would rise to five years in prison and a BDT 2 million fine (roughly USD 16,131), and to ten years and Tk 4 million if the target is a woman or a person under 18. A new Section 26A would punish publishing or spreading a “rumor” or “disinformation” with up to ten years in prison, a fine of BDT 4 million (roughly USD 32,262), or both. The draft defines a rumor as unverified information that causes or risks causing confusion, panic or social instability.
The government presents the changes as a response to harm already being done. Introducing an earlier amendment in June, the ICT minister said it would require platforms such as Meta to remove disputed content within a set time. Officials told Prothom Alo that the tougher penalties are meant to curb fake videos, audio and images made with AI.
Free speech concerns
Critics say the draft goes well beyond that aim. Transparency International Bangladesh (TIB) warned that terms such as “rumor” and “demeaning” are defined loosely enough to allow selective enforcement. A joint statement by rights groups said the draft mirrors the repealed Digital Security Act. The Committee to Protect Journalists has objected to the prospect of journalists facing criminal liability for reporting officials consider unverified. An Asia Times analysis noted that because defamation sits inside Section 25, the heavier penalty for offenses against women would apply to it as well, and not only to sexual abuse.
Numbers, gaps and local responses
What the law would address is substantial. A report by Dhaka Tribune said police's Cyber Support for Women unit had been approached by 60,808 women by September 2024, and cited a 2024 Bangladesh Bureau of Statistics–UNFPA survey in which nearly 89 percent of women social media users reported online violence at least once. Officials say the real number is likely higher because of underreporting. The published figures do not separate cases involving AI.
Researchers describe a similar pattern across the region. A study in the Journal of Online Trust and Safety drew on three workshops with 49 journalists, lawyers, fact-checkers and advocates in Sri Lanka, India and Bangladesh. It found that women are disproportionately targeted, that platform moderation often fails in South Asian languages, and that victims struggle to reach institutional help. Sithi's (a student activist known for her role in the historic student-led uprising in 2024) experience, where the cost of private help was out of reach, fits that last finding.
Digital skills are an added concern. The Asian Development Bank reports that functional digital literacy in Bangladesh stands at 49 percent and that rural users are 29 percent less likely than urban users to use mobile internet. Bangladesh Bureau of Statistics data released in April show that 56.6 percent of men use the internet, against 50.2 percent of women. Advocates argue that people with fewer digital skills may find it harder to spot manipulated media or seek redress. These surveys, however, do not measure exposure to deepfakes directly.
Outside the government, fact-checkers are trying to keep pace. FactFest 2026, held on 27–28 September, brought together around 100 fact-checkers, journalists and researchers, according to an organizer. Its 13 sessions covered AI and gendered disinformation, among other topics. In the closing roundtable, speakers said Bangladeshi fact-checkers face shrinking funding and weak coordination with mainstream newsrooms.
The cabinet committee has yet to finalize the draft, which would then go to Parliament. Constitution Watchdog, a civil society group, has urged the government to withdraw the draft in its present form and reopen public consultation for at least 30 days. It also acknowledged the government's stated aim of protecting women and children from online abuse and responding to deepfakes. Whether a narrower law can do that without the speech provisions now under attack is the question facing lawmakers.
