Gender-based violence

The violence left out of the record

Reading time: 15 min

Who decides what counts as digital violence against women in twenty African countries?

·AIthropology Lab
Add AIthropology Lab as a preferred source on Google
Group photo in Lusaka with signs calling for participation in shaping the internet.
Participants at Decolonizing the Internet East Africa, a gathering of East African feminists organised by the African women’s network FEMNET and the Whose Knowledge? campaign in Lusaka, Zambia, on 26 and 27 September 2022. The signs call for reclaiming and rebuilding the internet ‘with and for us all’ and recall the slogan ‘nothing about us without us’. The photograph predates the study and is unrelated to it or the cases it documents. Image resized. · Namukolo Siyumbwa (Eness Photography) / Whose Knowledge? · CC BY-SA 2.0

On 3 November, the African Commission on Human and Peoples’ Rights published the draft of the first continental study on digital violence against women for consultation and scheduled an online session for Wednesday 11 November with the two rapporteurs responsible for the study. It reviews twenty countries and finds that none has a specific, comprehensive, gender-responsive law covering this violence as a whole. Hardly any measure it systematically, and none has a dedicated legal response to violence perpetrated through AI-generated content.

The draft is titled Digital Violence Against Women in Africa. A Continental Study. It was commissioned by the African Commission, the African Union body that monitors compliance with the African Charter on Human and Peoples’ Rights, and produced by the University of Pretoria’s Centre for Human Rights and Equality Now. According to the document, Sigi Waigumo Mwanzia, founder of the Digital Rights and Freedoms Regional Hub, compiled it from five regional reports. It was reviewed by Commissioners Janet Ramatoulie Sallah-Njie and Ourveena Geereesha Topsy-Sonoo, the Commission’s rapporteurs on women’s rights and on freedom of expression and access to information, respectively.

The study grew out of two Commission resolutions. Resolution 522, adopted in 2022, called on states to expand their definitions of gender-based violence to include digital violence, and Resolution 591, adopted in 2024, commissioned the study. A third, Resolution 665 of 31 July 2026, extended the deadline for completing it by a year. Written submissions of up to five pages are accepted until 4 December in English, French or Portuguese. One of the six questions in the call asks whether issues relating to AI-generated content, deepfakes and emerging technologies need more attention.

A study built from reports

The method shapes what the study can say, and the document is frank about it. It synthesises five regional reports that combined analysis of laws, court rulings and literature with interviews with organisations, public officials and internet service providers, and, in some regions, with focus groups and surveys of survivors. The continental study does not publish the primary data from these reports. It covers 2019 to June 2026 and selects four countries per region. Central Africa includes Cameroon, Chad, the Democratic Republic of the Congo and Gabon. East Africa includes Ethiopia, Kenya, Rwanda and mainland Tanzania. North Africa includes Algeria, Egypt, Morocco and Tunisia. Southern Africa includes Botswana, Mauritius, Mozambique and South Africa. West Africa includes The Gambia, Ghana, Nigeria and Senegal. These are twenty of the African Union’s fifty-five member states.

The team cautions that official, disaggregated data on digital violence are unavailable in many of these countries. The figures indicate the scale of the problem and should not be pooled into a continental estimate. In its conclusions, the draft says that approximately one in three women surveyed or interviewed had experienced digital violence, although it does not explain how it derived that proportion from studies with different samples and questions. The figure cannot be used to estimate how many women across the continent experience it. In North Africa, estimates range from 11% to 89%, a spread the study attributes to incompatible national methodologies.

The absence of data is itself a finding. None of the four East African countries has a national database on gender-based violence that includes digital violence, and cybercrime statistics in Kenya, Rwanda and Tanzania are not broken down by the sex of those affected. Mauritius is the only one of the twenty with a dedicated monitoring system, drawing on its Research and Innovation Council’s dashboard and its national computer incident response team. The picture has begun to change since the study closed its data collection. In early November, the Kenya National Bureau of Statistics presented its 2025–26 demographic and health survey, which for the first time includes a module on technology-facilitated gender-based violence. The published results say that 26% of women aged 15 to 49 have experienced it at least once, with husbands and partners among the most frequent perpetrators. On 3 November, ‘Contar a las que faltan’ drew on anthropologist Sally Engle Merry to recall that whoever defines an indicator decides what counts as a problem. The continental study describes the consequences of having no such indicator. Without a baseline, it says, governments cannot gauge the scale of the problem, allocate resources or measure progress. Underreporting deepens a data gap that makes it easier to deprioritise the issue.

The study builds on work that African organisations were already doing. In 2020, Uganda’s Pollicy published feminist research on online violence in five African countries. Much of what the draft synthesises comes from organisations, journalists and researchers who have documented it for years.

Laws that fail to see and laws that punish

Seventeen of the twenty countries have ratified both the Convention on the Elimination of All Forms of Discrimination against Women and the Maputo Protocol, the African treaty on women’s rights. According to the study, none has enacted specific, comprehensive legislation on digital violence against women. Tanzania alone has included it in a national plan, a step that has yet to lead to legislation. In all twenty countries, including Tanzania, survivors turn to laws on cybercrime, defamation or public morality that offer partial protection and were designed for other purposes.

The study highlights one effect of those laws. In the Democratic Republic of the Congo, Ethiopia, Kenya, Rwanda and Tanzania, provisions on morality, obscene publications or public decency can be applied as readily to the woman whose intimate image was shared without her consent as to the person who shared it, including when AI was used to fabricate or alter the image. The survivor risks experiencing the abuse and then being penalised under the law. These provisions can also be used against women’s own expression. The North African report cites Egyptian model Salma al-Shimy, who was sentenced at first instance in April 2023 to two years in prison and a fine over her own social media videos, on charges of ‘violating family values’. In November that year, an appeals court overturned the prison sentence and upheld the fine. The draft dates the conviction to 2024 and omits the appeal. In South Africa, the study describes alleged perpetrators using defamation suits and protection orders to silence people who report them. Pumla Dineo Gqola, whose concept of the female fear factory featured in Tuesday’s note on South Africa, said in a 2023 interview that her work after 2015, collected in Female Fear Factory (2021), extended to countersuits by convicted perpetrators.

Where laws exist, enforcement falls short. According to the study, police and judges often lack knowledge of applicable provisions or treat digital violence as a private or family matter. Digital evidence is lost, collected improperly or ruled inadmissible. In West Africa, the data cited by the study show that only 23% of people affected report it, and 78% of those who do not report say they have no confidence in the outcome.

Platforms, evidence and languages

The study documents AI-generated sexual images in all five regions, mainly targeting women in public life. None of the twenty countries has rules designed for this abuse. None of the four East African countries addresses synthetic content in its rules of evidence, and survivors say evidentiary difficulties prevent them from obtaining redress. The document mentions Grok, X’s AI assistant, and says users have used it to generate images of naked women and girls. According to the draft, this use remains largely without specific regulation in the twenty countries. In January 2026, following criticism and investigations in several countries, X announced that it would prevent Grok from editing images of real people to show them in revealing clothing and would block generation of such images in jurisdictions where it is illegal, a condition that puts the question back in the hands of national lawmakers.

The study assigns responsibility to companies and governments. According to the draft, platforms invest far less in safety in African markets than in those that bring in more revenue. They moderate content in local languages poorly or leave it unmoderated, and amplify material that drives engagement even when it causes harm. In West Africa, according to data cited in the study, they take an average of 72 hours to respond to a report of online gender-based violence. Governments do little to hold them to account and, in several of the countries studied, appear more willing to regulate online dissent than to curb violence against women. Recommendations include mandatory investment in local-language moderation, maximum response times, assessments of the harm algorithms cause women and expedited routes for taking content down.

Language is a barrier that often goes unnoticed, the study says. In Senegal, 87% of digital safety materials are available only in French, leaving out an estimated 38% of social media users who struggle to read it. The document concludes that the design of support systems can reproduce the inequalities they are meant to address and calls for information to be available in the languages people use every day. The consultation has a similar limitation. The call was also published in Arabic, but submissions are accepted only in English, French and Portuguese, even though the study covers five countries where Arabic is an official language, the four North African countries and Chad, and the African Union also recognises Swahili and Spanish as official languages.

Three episodes in Kenya

Three recent episodes in Kenya illustrate some of the tensions the study describes. On 13 March, Africa Check debunked AI-generated images and a video depicting Gladys Wanga, the governor of Homa Bay, in a fabricated situation at a religious event. Before April, a fake intimate video purporting to show Senator Karen Nyamu circulated on several platforms. According to journalist Maryann Muganda, writing in The Standard, it was made from a 2024 photograph. Nyamu sponsored Kenya’s first AI bill, which included prison sentences for harmful deepfakes. In a report that became public in October, the Senate’s information technology committee recommended that the bill should not proceed because it considered the fines and obligations too onerous for a nascent sector. The full Senate has yet to vote on it, and Kenya still has no dedicated AI law. A response may come through another route. A committee of the National Council on the Administration of Justice, chaired by Christine Kungu of the Federation of Women Lawyers in Kenya (FIDA-Kenya), has proposed amending the 2006 Sexual Offences Act to criminalise, among other conduct, the creation and distribution of AI-generated or AI-altered sexual images. The proposal still has to go through Parliament.

The third episode falls outside the study. In February, the Swedish newspapers Svenska Dagbladet and Göteborgs-Posten, drawing on interviews with more than thirty workers, reported that staff at Sama in Nairobi, a contractor that labelled data for Meta, reviewed recordings from Meta’s smart glasses to train its systems. The workers said the material included intimate scenes recorded in homes, including one in which a woman apparently did not know she was being filmed. In April, after losing the Meta contract, Sama announced it was laying off 1,108 workers. The footage thus reached outsourced teams in Nairobi, who reviewed intimate scenes involving people who may have been unaware that anyone would see them. The workers brought the case to light by speaking about it.

What the definition leaves out

A continental study can give the impression that it describes a uniform reality, and the team expressly guards against this. The twenty countries differ greatly in legal systems, languages, connectivity and conflicts, and the text insists that its conclusions cannot be applied uniformly across all states. It finds recurring mechanisms in these very different contexts. Digital violence grows out of the same patriarchal norms as offline violence, and the internet makes it faster, more anonymous and harder to erase. The study documents cyberharassment preceding femicides in Kenya and Tanzania, women forced to move in Ethiopia and South Africa after their addresses were published, and female students in Mozambique driven into silence by intimate images. It therefore describes a continuum between the digital and the physical and argues that laws treating them as separate phenomena fail survivors.

The definition also determines who is included. The study says it follows Resolution 591, examines violence directed against women and girls, and expressly excludes trans, non-binary and gender-nonconforming people from its definition. This leaves trans women outside an investigation devoted to violence against women, although the study itself recognises that LGBTI people are among those most exposed. Data are also virtually absent on women with disabilities, migrant and refugee women, and women living in conflict zones. On 9 November, this series examined the attack on trans women and travestis in Buenos Aires. The first continental study on this violence has had to decide, like any count, whom it includes. Its final version will show whether the consultation changes that decision.

What comes next

The Commission expects to present a final version at a later ordinary session after considering submissions. Its recommendations include comprehensive laws developed with survivors and women’s organisations, repeal of provisions that criminalise survivors, training for police and judges, disaggregated data, information in local languages and enforceable obligations for platforms. Part of the regional framework is already in place. The African Union Convention on Ending Violence Against Women and Girls, adopted in February 2025, covers cyberspace. It needs fifteen ratifications to enter into force, and as of early November only The Gambia and Ghana were recorded as having ratified it. The United Nations’ 16 Days of Activism begin on 25 November, and the Commission is accepting submissions until 4 December.

This article was written in Spain, drawing on the draft study, the African Commission’s call for submissions, and Kenyan, South African and Swedish press coverage. The article draws on the draft circulated ahead of the 11 November consultation. We did not interview the study team or the women cited. The draft may change after the consultation. The figures are from the version published on 3 November.

Sources