Tracking gender-based violence in the Larger World
We worked with partners from Jordan, Kenya, and West Africa to show the unique contours of online harm in their communities.
We worked with partners from Jordan, Kenya, and West Africa to show the unique contours of online harm in their communities.
Since 2023, our team has run multiple studies focusing on gender-based violence online, covering everything from sexual harassment to doxxing to other coordinated forms of abuse. This work has naturally focused on harms faced by communities in the Larger World, where the majority of our partners work and live, and where tech-facilitated gender-based violence (TFGBV) has received significantly less attention from both industry and the research community.
In our latest study, we built datasets of TFGBV drawn from the social web in order to develop machine learning classifiers that could detect (and ultimately reduce the spread of) this kind of content in Swahili, Levantine Arabic and African French. We documented familiar trends ranging from “morality”-based takedowns to sexualized attacks intended to silence female political activists.
The high volume of posts in our dataset that contained a mix of languages and local slang also forced us to grapple with the ever-changing nature of how we communicate. While formal grammar and style tend to remain static, the language that people actually use in everyday conversation is forever in flux. This is profoundly true in regions of the Larger World where our study was rooted.
Consider Sheng, a constantly evolving mix of Swahili, English, and other languages spoken in Kenya that varies “across age groups, social class, geography, and even platforms.” As we write in the post, in the everyday use of Sheng, “new words and expressions are invented on a near-daily basis. A classifier trained on yesterday’s vocabulary might miss today’s nuance.”
This poses an intractable problem for technology. Even with the best of intentions, machine learning tools are simply not designed to keep pace with dynamics like these. That doesn’t mean we can’t use them to better detect and reduce the spread of harmful speech online, but it does mean that there are limits to which problems classifiers can solve, and which ones lie closer to the roots of gender-based violence — a social and ultimately very human problem.
We’re proud to have conducted this project hand-in-hand with three regional partners — the Jordan Open Source Association, the Kenya Internet and Communications Technology Action Network (KICTANet), and HerSafeSpace (a program from the Nigeria-based organization Brain Builders Youth Development Initiative) — and we’ll look forward to publishing our research results later this year.
We highly recommend Eib (“Shame” in Arabic), a new Arabic language podcast from our friends at Sowt. Each episode of Eib is grounded in real experiences and perspectives that reflect how online violence, harassment, and gendered disinformation are felt, normalized, and too often dismissed in daily digital interactions. We’re also proud that Meedan’s research, data insights, and contextual expertise have been woven into episodes at key moments to help listeners make sense of what they’re hearing. Give it a listen.
We recently signed onto a public letter, organized by AccessNow, calling on technology companies to stop providing AI systems for use in military kill chains, and calling on states to stop using these systems. The letter states: “AI-accelerated warfare is rapidly becoming a means of rubber-stamping killing at speed and at scale, and currently no technical or procedural fixes can effectively prevent the lethal and devastating consequences that stem from the fundamental challenges it poses to international law.” We urge our allies to read the letter and share it widely.
This week, we released our latest annual report, which details our engineering achievements, programmatic work, strategic priorities, and financials for the 2025 calendar year. Have a look at this report, and more than ten years of annual reports on our website.
Contact us to explore collaboration opportunities.
With all what AI is doing to the world, why exactly are we building another damn chatbot?
Meedan Executive Director Dima Saber has pounded the proverbial pavement of the global tech-for-good conference circuit this season, making the case for why the world needs Suwali, our new chatbot for media and civil society organizations. Last week, she put her thoughts down in a fresh new essay for the Suwali blog. Here’s an excerpt:
When an AI speaks your language, is grounded in knowledge you trust, and has the discipline to say “I don’t know” rather than invent an answer — that is infrastructure. Infrastructure that helps communities resist misinformation, allows organizations to re-engage their audiences, offers people reliable answers to questions they were afraid to ask out loud.
Does the world need another chatbot? Not really.
Does it need AI that is built with communities rather than deployed on them, that speaks their languages, respects their privacy, and isn’t designed to extract their data to feed it to commercial tech?
Yes. Urgently. And we’re building it.
July 6
MataSEA invites individuals, collectives and groups to contribute their stories on experiences of technology-facilitated gender-based violence (TFGBV) in Southeast Asia for the first-ever issue of MataZine. Submit by July 6.
July 8-9
SRCCON (pronounced “source con”) is a peer-led conference from OpenNews for journalists who want to transform their work, their organizations, and their communities. This year’s edition will take place in Minneapolis, U.S. Registration is open now.
September 4-6
The Global Gathering will once again take place in Estoril, Portugal this September, bringing together groups from around the world that are working on the most urgent technology-related challenges affecting human rights, social justice, civil society, and journalism at the local, regional and global levels. Apply by June 30.
(De)monetizing Repeat Disinformers: Meta's Monetization Enforcement in Practice
“We analyzed 290 Facebook pages fact checked 10+ times by one of Meta’s Third-Party Fact-Checking partner between 2020 and 2025 and found that Meta’s existing mitigation measures fail to prevent repeat disinformation offenders from accessing monetization programs.”
(What to Fix)
Abortion Access Denied: Investigating the Global Censorship of Women on Web
“Censorship of reproductive rights information is on the rise globally….Our latest analysis revealed the blocking of Women on Web domains in Iran, Türkiye, the Philippines, South Korea, Spain, Kuwait, and Argentina. OONI data shows not only that access to Women on Web’s main website is restricted in a growing number of countries, but also that the blocking is more widespread and consistently observed.”
(Women on Web and Open Observatory of Network Interference)
How AI’s Labor Supply Chains Fail Workers
“Access to labor is streamlined but clients never have to think about who is actually doing the work. Annotators, labelers, and trainers — most either from or based in the Global Majority countries — are largely invisible despite forming the foundation of the AI industry.”
(Marché Arends, Tech Policy Press)
