The following statement was issued today by the Spokesperson for UN Secretary-General María Teresa García:

The Secretary-General is deeply concerned by the continued escalation of AI syntactic warfare, most recently targeting institutions of the European Union in Brussels and the headquarters of the World AI Cooperation Organization in Shanghai, which has caused chaos among civilian populations worldwide and the reported displacement of 2,580,000 civilians.

She is alarmed by inflammatory rhetoric targeting institutions at the heart of multi-lateral co-operation for peace and security, including the United Nations, and the announcement of expanded AI syntactic-military operations, which will further harm civilian populations who are already in a vulnerable situation.

The Secretary-General calls on the Governments of the United States, China and Russia to take immediate and decisive action to halt all AI syntactic-military operations and de-escalate the situation through inclusive dialogue. The Secretary-General further calls on all governments to protect civilians from this cognitive warfare and to ensure access to safe, secure digital spaces away from AI syntactic agents.

None of what I just read is true. Maybe you’d already figured that out, but I could see at least some of you wondering: is Outlook down because there’s been another AI hack, and this time it’s really got out of hand?

Or maybe you were wondering: did I miss the news that we have a new UN SG who is a woman? Not sure which of those two you would find more surprising.

Build Up has been co-organizing the Build Peace conference series since 2014, and this year we are delighted to be working with the Kindred Center for Peace Advancement and the Global Center for Pluralism to bring the conference to Waterloo, and to have the support of the Balsillie School of International Affairs and the Council on Technology and Social Cohesion to make it happen.

On behalf of the whole organizing team, welcome to the 13th annual Build Peace conference!

And no, AI warfare hasn’t started.

What I read is actually the opening of a short sci-fi story that Julie Hawke and Helena Puig Larrauri wrote back in June. See the story here.

This conference isn’t about AI, but because we know it’s something on everyone’s minds and that will come up in many sessions and conversations, we want to take some time to share how we are thinking about it at Build Up. And that’s connected to this short story.

We wrote the story for two reasons.

First, we wanted to highlight that existential threats from AI are monopolizing our imagination about the future to the detriment of paying attention to what’s happening now, to the threats that are already here. We actually wrote it before the latest spate of hacking incidents by AI agents, and then as the months went by things kept happening that made it seem more real than sci-fi.

The imagined Global Crisis in the story isn’t the result of a super intelligence that wants to end humanity. It’s a near future crisis built on what we are seeing now. We describe it like this:

It wasn’t the AGI event that everyone feared. Crises are usually mundane, when you look at the patterns close enough. AI agents first were granted signatory authority on routine infrastructure contracts because they negotiate better terms faster. AI-run holding companies, named even earlier as legal persons, began owning infrastructure. Oversight to internet protocol governance bodies had already been severed. The agents optimizing simultaneously for war and profit simply saw the opportunities and secured root certificates at the optimal time to do so. Hardly anyone noticed until they tried to reverse it.

Doesn’t sound so out there, right?

The story then revolves around the aftermath of this mundane crisis, and how communities rebuild in the cracks. Like this conference, the story is about a rupture and the encounters that follow. We imagine a new form of governance, called the Residual, established in the tradition of the seed bank. Here’s how we imagine it:

The Residual is a multilateral body of communities who had never organized their knowing around distances or differences, and therefore resisted fragmentation. What AI syntactic technologies could not weaponize, it turned out, was knowledge that had never been made fully legible to them. As international cooperation began to falter, communities sought the continuity our network held. We didn’t replace the United Nations, though some asked us to, because we do not replace. Instead, we root and resource one another in shared and particular lifeworlds, and we re-plant wholeness in fertile cracks.

Although the story is fictional, it lays out the two things we are watching for, as AI enters more and more spaces: the consequences of threats and harms that are already here and the community responses that grow in the cracks.

Both of these are already showing up in our work, as we are sure they are showing up in yours, in ways that intersect with the three sub-themes of the conference: resisters and fixers, rural and urban, and origins and diaspora — across these divides, AI amplifies ruptures, and AI can support encounters.

How we are seeing in our work

We see this in the work we are currently doing in Bangladesh, the Netherlands, Kenya, Nigeria, Sudan, Somalia, and other parts of the world. But let me zoom in on two specific contexts: some of the work we are doing in Kenya — supporting peaceful elections ahead of the 2027 General and Presidential elections, and in Sudan, where we are engaging citizens in an active conflict situation.

Resisters and fixers are both needed, whether it’s an election cycle or in conflict.

The National Coalition for Freedom of Expression and Content Moderation is our resister posture in the Kenyan conversation. It is a network of stakeholders, including civil society, regulators and government institutions, media and international partners, who work together to push back on how social media platforms currently handle harmful online content, with automated AI moderation taking center stage, and asks who gets to decide what is acceptable in Kenyan online spaces.

Deliberative Tech platforms like Mersal and zKe are the fixers. Mersal began in Sudan to support online deliberation in a conflict setting, while zKe adapts that approach for Kenya’s 2027 elections, giving citizens a structured space to say what matters to them and see where others agree and disagree.

Resisting and fixing do not compete in this work. Resistance shows where the system fails, and the fix gives people a better alternative.

Political fault lines increasingly follow geography in Kenya and in Sudan. In Kenya, AI helps weaponize narratives that cast both rural and urban citizens as opponents with opposing interests. These narratives spread quickly because they reduce a complex economic picture to a simple us-versus-them story. In Sudan, the divide takes a different form. Conflict, displacement, and disrupted connectivity mean that people in different parts of the country experience the war very differently, and online conversation is often dominated by those who can stay connected. AI can widen that gap by amplifying the loudest voices and the most polarising content. It can also help close it.

In Kenya, we use AI to link online deliberation with in-person barazas in rural areas, so insights move between the digital process and conversations happening in physical rooms. In Sudan, Mersal gives citizens in an active conflict a way to take part in deliberation despite the constraints they face. In both settings, the aim is the same. People with limited access should shape the same conversation as those with the most.

Diasporas take part in the political life of both countries, and their narratives flow back into online spaces at home. In Kenya, diaspora voices carry credibility and reach, so they tend to shape how people at home read and respond to electoral conversations — this is why political leaders are doing a lot of diaspora convenings ahead of elections. In Sudan, the diaspora is a major presence in the conversation about the war. Many Sudanese abroad have family in the conflict, and they act as a source of information, advocacy, and support. This gives them influence, and it also means that narratives from outside can harden positions inside. AI contributes to this by amplifying and speeding up the translation, repackaging, and spread of content across audiences and platforms. It also lets manipulated content cross borders with little friction. In response, we work with influencers who hold the attention of audiences at home and abroad. The goal is not only narrative change, but to have those with reach model different ways of discussing political difference, so the diaspora connection becomes a bridge, and not a channel for division.

So that’s how the ruptures and encounters brought about by AI’s entry into the mass market are showing up for us.

But Helena said earlier that there were two reasons for writing the sci-fi story. The first was to draw attention to what is already here, and the possibilities growing in the cracks. The second was that we were trying to write Build Up’s AI policy and it wasn’t going so well.

We read many organizational AI policies, in our field and others, and were struck by two things. First, they were very abstract. “Keep the human in the loop” — sure, but what happens when that compromises AI’s promise of efficiency? How do we decide how much the human is in the loop? When, and what levers to use to counter presumed greater knowledge?

“Always be aware of bias in the data, and correct for it” — do we even have the ability to do that when even the best models available to us, always sycophantically nudge us to think we are already in the right?

The second thing we were struck by is that many policies read as protection from a future pre-determined by the inevitable forces of AI, as defending from something rather than co-creating a new thing. They read as security rather than peace.

We want something different for our AI policy.

At Build Up we approach technology as intrinsically entwined with society. We think conflict today is conditioned simultaneously by technology that is embedded in society and social relations that are immersed in technology.

So the question is not whether you chose to use AI, it’s that AI is changing who we are while all the while we have the agency to define what AI is. It’s a relational problem.

We wanted to write a set of principles that spelled out how we relate to AI, and from there write a policy that was actionable. These principles needed to be our moral imagination towards AI for peace, they needed to allow us to imagine a future that is not pre-determined by the inevitable forces of AI.

Operationally, they needed to put peacebuilding agency at the center. Julie wrote an article about that core idea where she explains that AI tools are most useful when they are embedded within and supporting people-centered peacebuilding processes, not the other way around. The “human in the AI loop” discourse of responsible AI emphasizes that human input and judgement is integrated at key stages of a model’s workflow. We advocate for an inversion of this principle: putting AI in the human loop.

Why this flip? When AI tools promise efficiency, scale, or synthetic aggregation, there are high costs for peace processes. Many uses of AI can turn participation into a proxy rather than a practice by replacing engagement with summaries, predictions, and synthetic representations. But we know dialogue processes work when participants feel heard, not accurately summarized in a generated report. So how can we imagine a use of AI that helps us listen better?

It is also technically possible, for example, to build a hate speech classification system trained on synthetically-generated data, governed by labeling rules created without community input, and validated primarily by large language models rather than affected groups. But we know early warning systems fail not for lack of data, but because the political will to act on warnings is absent. So how can we imagine a use of AI that builds political will?

We haven’t written our AI policies yet, so I don’t have a concrete answer to these questions.

For now, we just have the start of our principles for how we put AI in the peacebuilding loop. They are captured in the sci-fi story as a prompt that a UN mediator gives to an AI co-pilot that helps end the Global Crisis. And they are also our invitation to you for how to show up with one another throughout this conference. They go like this:

When the objective is to improve human interaction, don’t optimize for engagement, optimize for human agency.

When the objective is sense-making, don’t optimize on what is already digitized, optimize for discovery and connection.

When the objective is to tell a story, don’t optimize for efficiency, optimize for complexity.

When the objective is to provide decision making support, don’t optimize for neutrality, optimize for multipartiality.

And we’ve talked mainly about AI, but these principles apply to the use of many other digital technologies. We find it helps to both have general principles that apply to how we relate to technologies, and to be very specific about what technologies we are discussing in a conversation. It’s always better to say:

We need to understand how online discourse is entrenching and escalating the conflict. We are using supervised text classification to analyze content inciting violence on social media.

Instead of saying I’m using AI to build peace.

We need to make sense of a wide public input process. We are using pattern-detection tools to organize open-ended consultation responses so facilitators can engage with dominant concerns while still identifying outlier and minority views.

To say specific things like this rather than to say we are using technology to build peace.

And maybe these principles are relevant to how we relate to each other as well as how we relate to AI. Each of these objectives require imagination, and we invite you to look to both the future and the present with the openness and creativity that encounter asks.

Missing those who are not here

We wanted to close by sharing that there are at least 26 people who will not be able to join us because their visas were delayed or rejected. We also had an additional 11 initial visa rejections that were then approved on appeal. These are just the numbers for the people whose travel the organizing team supported, this doesn’t include other participants who were making their own arrangements. The rejections included Build Up’s Board Chair who is a Nigerian national, with permanent residency in Switzerland and teaching at a Swiss university. The delays included a Build Up team member who is also our conference videographer.

Of the thirteen conferences organized to date, this year has the most — an unprecedented number of — visa rejections and appeals. On behalf of the Build Up team and our co-organizers, I want to express our disappointment and solidarity with everyone who was rejected or had to bear the burden of additional scrutiny.

We did our best to select and prepare to host Build Peace in a visa-challenging location to maximize the likelihood of approval, but that has clearly proven insufficient. We know this has a significant negative impact on many participants and on the conference overall.

It’s not ok. For the next two years, we will not hold the conference in a country in the Global North. We may make that a permanent shift for all future conferences.

For now, we ask you to keep in mind the voices that are missing, as you move into the rest of the day.

How to follow along

For anyone not able to join us, Build Peace Radio is now live and will be broadcasting 24/7 through to the end of the conference, including all plenary sessions. Listen to it on the conference website. Click the radio button on the bottom right to listen live, see scheduled broadcasts and re-broadcasts, and chat with the radio team and other listeners.

This article was written and posted by the Kofi Annan Foundation, our partners for the film’s screening at the upcoming Geneva Peace Forum. Their original story here. Over the coming months we will be sharing more opportunities to join screenings, host your own, and to contribute to the ongoing conversations and actions that this film captures. If you want to be the first, join us in Geneva or online!

Across Ethiopia, Kenya, Nigeria, Somalia and Sudan, digital harms are reshaping how conflict unfolds. Ground Signal follows the practitioners, researchers and digital rights defenders doing the hard, essential work of pushing back.

Ground Signal is a 20-minute documentary by filmmaker Jean Marie Ndihokubwayo exploring the growing impact of digital harms in conflict-affected contexts across Africa — and the locally led efforts working to address them.

Disinformation, hate speech, online gender-based violence and the manipulation of information ecosystems are increasingly inseparable from how conflict unfolds, communities are polarised and democratic processes are disrupted.

About the film

In Ground Signal, Jean Marie meets digital peacebuilding practitioners working across Ethiopia, Kenya, Nigeria, Somalia and Sudan, uncovering the demanding but essential work they are doing and the emerging connections between efforts within and across countries.

Produced by Build Up in 2026, the film was created as part of an initiative to convene locally led alliances of civil society organisations addressing digital harms in conflict-affected contexts across West and East Africa.

Through interviews with practitioners, researchers and digital rights defenders, the film explores what digital harms look like on the ground, how local actors are responding, and why greater coordination — across organisations, sectors and countries — is essential to driving systemic change.

Documentary screening

Date: 14 October 2026
Location: CICG — Centre International de Conférences de Genève and online

Register for the screening event

Ground Signal will be screened by the Kofi Annan Foundation and Build Up as part of Geneva Peace Forum. The session will bring together perspectives from practitioners working to address digital harms in conflict-affected contexts across Africa.

The screening will be followed by a panel discussion and audience Q&A, bringing together the filmmaker and practitioners from the region to explore the challenges and opportunities for addressing digital harms.

The conversation will look at locally led and multistakeholder approaches to digital harms, social media analysis in conflict contexts, youth-led responses, and the role of collaboration in developing more effective responses.

The discussion also draws on findings from complementary research examining multistakeholder engagement in digital harms regulation and policy across Africa.

About the film

Title: Ground Signal
Director: Jean Marie Ndihokubwayo
Production: Build Up
Year: 2026
Duration: 20 minutes
Language: English
Subtitles: French, Arabic and Somali
Featured countries: Ethiopia, Kenya, Nigeria, Somalia and Sudan.

Crossposted from the Journal of Social Encounters, available here.
Hawke, Julie (2026) “Partial Coalitions and Peacemaking as Metaphor,” The Journal of Social Encounters: Vol. 10: Iss. 2, 247–250. DOI: https://doi.org/10.69755/2995-2212.1452

I watched with a particular curiosity as Magnifica Humanitas was shared and re-shared within my distinct and overlapping networks of peacebuilders, tech governance professionals, digital rights defenders, civic-tech builders, and techno-social researchers. I have worked at the intersection of technology and peacebuilding for over a decade and am a doctoral researcher at a private Catholic university. While having a general appreciation for Catholic social teaching, Catholic Worker Houses, and associations of friends and colleagues, I transparently did not know what an encyclical was until Francis’s Laudato Si (2015), and this only when engaged at the University of Notre Dame. I assume that many of the broad “people of good will” Magnifica Humanitas was directed towards and who participated in its spread were similarly surprised by its resonance. I include this positioning to foreground that my commentary will address the substantive topics of AI as it intersects with violence and peacemaking, but also to explore the awkward position entailed in subscribing to and promoting a doctrinal resource for a powerful political actor who simultaneously impedes reproductive rights, LGBTQ+ inclusion, and women’s ecclesiastical equality. In what follows, I will include comments on the content and the circumstances of its publication.

Internal and external resistance

The encyclical itself arrived wrapped in its own contradiction. The partnership with Anthropic, whose business model the letter explicitly indicts, rightfully surfaced critiques of moral laundering. Accusations of “vatican-washing” or “pope-washing” observe that proximity to the papacy provides legitimacy and distraction from data theft, defense contracting, and labor and resource exploitation (Jenkins, 2026). At the same time, the letter’s comprehensive account of technological power leaves no ethical stone unturned. It names the technocratic paradigm continuous with “wars, colonialism, racial or gender discrimination, violence against entire peoples and exploitation” (para. 79). It privileges the rights of workers, the human consequences of consumption and supply chains, and argues that disarming AI is a task “ecological in the deepest sense” (para. 110). It covers surveillance, algorithmic bias, monopolies, and data sovereignty. I believe it is safe to say that Magnifica Humanitas is the most serious statement yet on the harms of AI from the leadership of an institution with over one billion members. The temptation is to resolve the dissonance with a stance. The document is compromised by its company, or the critics are compromised by their politics.

When the technocratic paradigm is outlined as “dominant,” I read hegemonic and interpret through Gramsci’s (1971) lens of an unsettled, continuous process of negotiation and persuasion. Gramsci understood that any bloc capable of both maintaining or contesting a hegemonic order is assembled from misaligned actors sharing a conjuncture rather than a worldview, and blocs are formed and maintained through the formations of strategic, fragile, and interest-aligned coalitions. It is worth a longer exploration than this commentary allows that Gramsci developed these theories while imprisoned by a regime that the Catholic Church helped legitimize for mutual benefit, using the Church as the exemplar for a war of position. A less generous reading focused on the circumstances of the publication sees another elite pact between culture and capital.

However, Magnifica Humanitas also supplies its own image of this contest. The encyclical frames peacebuilding in the age of AI as a choice between two construction sites–the Tower of Babel and the rebuilding of the walls of Jerusalem–to represent two ways of organizing a shared world. Jerusalem is Pope Leo’s counter-hegemonic answer to Babel, where walls of “fraternal coexistence” are raised by working together without discarding difference. In the encyclical’scontent, we read advocacy for regulation, for workers, for restitution, reparation, justice, for peace as a norm, and against surveillance, exploitation, and wealth inequality. The content makes no concessions to the self-sustaining interests of power, and the program it lays out provides enormous cultural capital for the bloc resisting these new, broad formations of harm.

I am now getting to the work of peacebuilding, which as a field, for better and worse, accepts partial participation and assumes compromised actors as a default. In the encyclical, Nehemiah’s story is uplifted as our guide, where among other things, he “confronted internal and external resistance” to build the wall. While not further explained there, the biblical story details people building with tools in one hand and a weapon in the other and not only resistance but exploitation within. Wealthier workers leveraged the famine to seize the fields and even the children of the poor as debt bondsmen, and Nehemiah paused the building in order to force restitution. I will not pretend to be skilled at exegesis, but a story of a compromised alliance building and contesting itself at the same time resonates more than a pure bloc set against a corrupt one. I agree with the encyclical almost entirely in the diagnosis of the ruins we are standing together in, or the socio-technical conjuncture undergirding new industries of overt and covert violence. My concern is that naming the work begins to feel like doing it when building is used as a metaphor instead of a blueprint.

Blessing, and equipping, peacemakers

Pope Leo aptly observes that peacebuilding has been “relegated to a secondary role,” with cooperation, disarmament, conflict prevention, and trust-building neglected in the name of power politics (para. 203). An argument for peace is given central attention in the letter, yet in terms of specification, the letter partly reproduces the relegation it laments. The workings of war and weapons receive operational criteria and non-negotiables (para. 197; para. 200). In comparison, dialogue, diplomacy, and forgiveness are invoked as humanity’s superior tools (para. 192), but the treatment remains institutional and dispositional, rather than operational. The United Nations is an “essential instrument” and mercy is a “concrete criterion” (para. 226–227). Only once, where diplomacy is charged with “negotiating shared regulations on the use of digital technologies”(para. 225), does the peacemaking vocabulary approach the technical specificity granted to war. Having affirmed that innovation can serve the common good, with particular promises of a disarmed AI, what is possible for peacemakers in the present?

The letter outlines five paths for individual and collective responsibility: “the need to disarm words, building peace through justice, adopting the perspective of victims, cultivating a healthy realism and reviving dialogue and multilateralism.” (para. 213). Each of these have digital and technological counterparts peacemakers should attend to but are not named in the letter. We can specify some now. Disarming words includes the personal commitments outlined but also requires privileging or creating platforms that incentivize prosocial behavior and penalizing platforms that foster and profit from hostility. Individuals have agency but act within environments. Practicing justice can involve equipping peacemakers to make injustice maximally legible with documentation and archiving or relocating the power to define harm by connecting affected communities. Adopting the perspective of victims also includes amplifying their voices and defending against their suppression and censorship, especially when the policies surrounding ‘disarmed words’ are used to silence dissent. Cultivating a healthy realism, which the letter defines as understanding ‘what can be achieved’ (para. 218), in practice is data work. Data and peacemaking requires and enables these kinds of assessments about operational opportunities and constraints as seen through conflict analyses, narrative assessment, monitoring tools and more. It also requires treating data as neither neutral nor sufficient and the recognition that the designation of data excludes knowledge that is not digitized. Finally, reviving dialogue and multilateralism also means lowering the barriers of geography, language, literacy, and cost that have structured participation in the past. Deliberative technologies offer opportunities to move beyond siloed, elite processes, and technology-facilitated approaches such as public consultations, specific interest group participation, decentralized deliberations, or participatory documentation can broaden inclusion and create avenues for marginalized or disempowered groups to advocate for their needs (Hawke, 2024).

Each of these sampled counterparts are sourced from active ongoing work by peacebuilders, so they do not lack viability. However, they do often lack the recognition and resourcing required to shift peacemaking work from conviction, where it is important, to building shared conditions and conducive environments, where it is most impactful. In short, peacemaking needs the cultural capital of a blessing, but it also needs equipping. In today’s construction site, that invites new specifications: alternative technical infrastructures like community-governed datasets, classifiers trained with affected populations rather than for them, or peacemaking tools hosted as digital public goods; interoperable dialogue and deliberation platforms; and pooled compute and technical capacities.

Peacemaking also needs increased alignment between internal operational cultures and externally applied mechanisms and methodologies. The letter was right to observe that technology is not inherently neutral (para. 9), and that those who use it take on its characteristics in turn. A perspective of tech neutrality easily slips into tech solutionism, which treats tools as key to solving political or relational problems dependent on the intentions of the deployers. Equipping peacemakers should not be considered as an outward-facing endeavor. An institution aiming to facilitate democratic, inclusive, or transparent peace processes externally using digital tools will lack the authority and experience needed if their internal digital integration is characterized by opaque, top-down, or exclusionary policies and procedures. The process of turning metaphors about the digital era into plans of how to work within a new environment is an opportunity to change and be changed by design. Crafting guidelines around data privacy, information security, and cumulative workload equity forces an institution to actively mediate competing internal interests, address grievances, and operationalize consensus-building frameworks. The risk of the alternative is that intentions do not fundamentally change practices, and safeguarding is actually surveillance, inclusion becomes extraction, and dialogue is synthetic.

Conclusion

Magnifica Humanitas names the ruins in AI’s wake with both breadth and precision, and in doing, it gathered an unlikely readership. The encyclical’s contradictions — indicting the business model of its distribution partner and its issuing from a politically engaged institution with its own exclusions — are a condition of the work of building together. Read through Gramsci and Nehemiah alike, coalitions capable of contesting a dominant order are contesting themselves even as they cooperate. However, assent to a metaphor costs nothing. Whether the readership becomes a coalition of builders rather than good-willed individuals depends largely on what the letter leaves unspecified: tools, infrastructure, and resources for turning techno-social peacemaking into practice. Each of the five paths for responsibility have digital counterparts already in practice and their further needs identified. Recognizing and naming these with the same operational specificity afforded to warfare would help peacemakers build from foundations already laid. If the letter is to remain a blessing, it is a very good one. It remains to be seen whether and how it could be more.

References

Francis. (2015). Laudato si’: On care for our common home [Encyclical letter]. Libreria Editrice Vaticana. https://www.vatican.va/content/francesco/en/encyclicals/documents/papafrancesco_20150524_enciclica-laudato-si.html

Gramsci, A. (1971). Selections from the prison notebooks (Q. Hoare & G. Nowell Smith, Eds. & Trans.). International Publishers.

Hawke, J. (2024). Digital inclusion in peacemaking: Practice, promise and perils. Accord 30. Conciliation Resources.

Jenkins, J. (2026, May 22). Inside the unlikely Vatican-Anthropic relationship that’s reshaping the AI ethics debate. Religion News Service. https://religionnews.com/2026/05/22/why-anthropic-is-helping-unveil-the-popes-new-encyclical-on-ai/

Leo XIV. (2026). Magnifica humanitas [Encyclical letter]. Libreria Editrice Vaticana. https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html

Over the past year, Build Up has convened a group of Sudanese civil society digital experts to monitor narratives on social media about the war. Our monthly reports track trends in narratives and their impact on offline events. Some of the recurrent narratives identified in these monitoring reports require a deeper analysis to unpack how they emerge, how they evolve, and their impact on offline violence.

The five thematic research reports discussed at this event offer this depth of analysis, and provide an important input to humanitarian and mediation practitioners working towards a peaceful Sudan. Inputs from the discussion at this event will be incorporated in the final reports, to be published later in the summer. If you missed the event, you can comment your feedback here or email us so they can be incorporated.

This event was organized by Build Up with the support of the UK’s Foreign, Commonwealth and Development Office, as part of our collaboration to address digital harms in conflict. With thanks to Conciliation Resources for hosting us!

A note for readers: the discussion references massacres and threats of violence in Sudan.

Do the armed actors have a deliberate communication strategy, and is it organized or paid for?

The coordination is often visible in the data. Twenty accounts will post the same content one or two minutes apart, which only makes sense if the activity is directed. Around major incidents, accounts that are not Sudanese get mobilized to amplify particular narratives and to justify violence happening offline. A clear example is the way the El Fasher massacre was reframed through engineered hashtags, with the language shifted from massacre to “liberation.” This is amplification working as part of the armed effort, where posts reinforce what is happening on the ground.

Are there spaces of hope on social media?

Yes, in specific places. Women-only and closed groups host real conversations, including across disagreement, where people who hold opposing positions still talk to each other. Civilians also began mobilizing differently around the third anniversary of the war on 15 April, trying new tactics and agreeing on shared strategy while letting different groups produce their own content. That moment faded over the following months. A recurring obstacle is that civilian voices often adopt the same language used by the warring parties, which undercuts attempts to shift the narrative.

This work means reading distressing material. How is that handled?

Researcher wellbeing is a real concern. The trauma of sustained exposure to violent content is common among social media researchers working on Sudan and elsewhere, and it is something the team attends to internally.

How do the platforms differ in their role, for example Meta, TikTok, and X?

The differences are mainly about community rules and enforcement rather than competition for users. Facebook designated the RSF as a dangerous entity early in the war and acted on related content, including explicit threats of sexual violence that were later taken down. X has little meaningful enforcement and is difficult to engage, and its tools can surface harmful content rather than limit it. TikTok amplified extreme material, including a live broadcast in which a perpetrator boasted about mass killings without the platform removing it, and is only now beginning to respond. A further gap is annotation. The companies lack Sudanese annotators, and word meanings shift quickly across communities and over the course of the conflict.

Sudan has millions of cellular connections and internet users. Are these figures built into the analysis?

The numbers matter, but they need translation. A count of connections is not a count of people, since one person may hold several SIM cards or accounts. Connectivity is also uneven, with some areas cut off from networks and relying on Starlink or other routes. The diaspora shapes the information war more than many users inside the country, partly because those posting from abroad face fewer consequences for what they share.

What methods are used, and where does AI fit? Have offline approaches to studying the actors been tried?

Offline research into actor incentives is not yet systematic, though there have been some conversations with influencers about what drives them, and this would strengthen the work. On analysis, the Phoenix platform uses keyword and lexicon classifiers as a living aid for analysts, alongside a fine-tuned classifier built for this data. Annotating the data to train that classifier is itself a peacebuilding process, since it requires people to agree on categories. The framing the team prefers is AI inside the peacebuilding loop, with people kept central and the tools assisting. The sheer volume of material makes some automation necessary.

How effective are offline versus online interventions at preventing these narratives?

There are few good systematic evaluations. Digital literacy is hard to make stick, and simply making people aware that manipulation is happening does little to change behavior. Targeting specific behaviors is more effective, which is why the team is interested in the incentives of high-impact influencers. This remains an area that needs more research.

Has any of this come up in peace talks, given that parties often have to rehumanize each other before negotiating?

Online hate speech comes up constantly in conversations with civilian and political actors, and the UN envoy’s team has flagged it as a concern. The thinking behind digital de-escalation is that parties will struggle to come to the table constructively if the online environment has not been calmed at least somewhat. Actors themselves say it will not stop until the war stops, but de-escalation can function as a confidence-building measure that makes talks more possible.

Are new or alternative platforms emerging, and what comes next?

Three systemic levers guide the work. The first is shifting the norms for how people behave on existing platforms, which is why influencer work matters. The second is alternative spaces, where the more practical approach is to meet people where they already are and move them into a more consultative environment, as with the Mersal initiative, rather than expecting wholesale migration to a new platform. The third is influencing platform and technology companies toward conflict-sensitive design. On automated warfare, target review times measured in seconds raise real questions about how far automation has already arrived.

There is a shortage of skilled journalists after so many were displaced. How does that factor in, and is Build Up working with local media?

The capacity gap is real and understandable given the scale of displacement and the loss of funding. Many journalists fled and now have to manage daily survival, which makes sustained reporting work difficult to expect of them. Supporting and funding this work is essential, and while some organizations are doing good work with journalists, more is needed.

Has Build Up engaged in advocacy toward the platforms, and what response came back?

Build Up is a trusted partner of Meta and a trusted flagger for TikTok, while dialogue with X is limited. The current focus is convening Sudanese partners to define a local approach to digital harms and a minimum standard for escalation, so that any ask to the companies comes from that base. One broader idea, developed with colleagues at the Neely Centre, is to press platforms to treat the period around peace processes and ceasefires the way they treat elections, when they already deploy break-the-glass measures.

We had more questions than time during the session! Themes raised but not addressed here included discrimination against Arab communities and the language used toward western tribes, comparisons with the Darfur conflict of the mid-2000s before social media, the persuasive capabilities of large language models on the battlefield, the public’s own adoption of warring-party narratives, and methods for distinguishing accounts run from inside and outside Sudan. We appreciate all of the feedback and additional lines of inquiry, and are taking them all on board as we prepare to publish these reports.

On Wednesday April 15th, foreign ministers, UN officials, civil society organisations, and key Sudanese actors convened in Berlin to discuss ways to de-escalate the war in Sudan and strengthen support for humanitarian response efforts. As a result of this meeting, the Diplomatic Service of the European Union published a communiqué detailing the process, objective, and outcomes of the conference, and participants launched a joint call to End the War and Advance a Sudanese-Owned Political Process.

In international conferences, particularly those addressing contexts of war and conflict, there’s always a struggle to include the local voices of those who are most affected and involved on the ground. In similar processes to the Berlin meeting, international and local peacebuilding organizations typically run consultations with different local actors and underrepresented groups, in an effort to share these voices with conference organisers and other actors in the room. However, such consultations are limited — because they tend to be in person, they are costly, logistically difficult to organise, and often inaccessible to many of the communities most impacted by the crises and struggles that such conferences tackle.

In an effort to address these limitations for the Berlin meeting, Build Up began coordinating with the UN Office of the Personal Envoy for Sudan as well as the EU Office of the Special Envoy to Sudan to reflect the opinions of Sudanese people via Mersal, a digital consultation platform created by Build Up in 2024 in an effort to form a space for collective and democratic engagement between participants. In collaboration with the Special Envoy and conference attendees, the findings and insights from the 1600 Sudanese people who shared their views on Mersal were used as input in the civilian meeting that launched the call for de-escalation.

Insights from 1600 Sudanese people who shared their views on Mersal were included in the Berlin conference civilian meeting that launched the call for de-escalation.

So, what is Mersal? How did it lead us to these results? Who participated? Why is it a consultation and not a survey? What is the importance of such deliberative processes? Can this replace in-person consultations? In this article, I try to answer some of these questions in an attempt to reflect on and document this process, sharing it with others who’d like to engage in this exploration.

What is Mersal?

Mersal is a digital platform for public participation. It provides an anonymous space for Sudanese people to engage with political and social issues, especially for those often marginalised in formal processes. Mersal, launched in May 2024, functions through a chatbot on WhatsApp, and currently has over 9k subscribers. Since its inception, Mersal has created a digital consultation space for the Sudanese public, where people can discuss questions about the social and political issues that affect them, share opinions, vote on pertinent topics, and view this content and its results publicly (see for example here, here and here). Mersal is not a survey tool; it’s a deliberation platform, allowing its users to interact with each other in a facilitated manner. Mersal currently operates on Polis, a platform that I will talk about later in this article.

How did Mersal start?

In May 2024, a group of Sudanese civilian actors came together to form Tagadom, a Sudanese coalition that was formed with the demand and the objective to end the war in Sudan. The founding conference Tagadom took place in Addis Ababa that month, with the aim of representing different segments of Sudanese society. While the participation in Addis Ababa was significant, with 600 people present in person, there was a need to also include participants who could not be physically present for a variety of logistical reasons and challenges.

At the time, Build Up was working on providing digital support for the convention. It was this process that led us to create the WhatsApp chatbot Mersal. Through our networks on the ground as well as with civil society organisations and the coalition members, we reached out to run consultations around the meeting itself. We asked questions about the representation within the conference, the humanitarian priorities and Tagadom’s role in this, and the priorities for the final output of the convention, and we also gained feedback on Tagadom’s principles for a political solution in Sudan. Mersal was launched on May 21st just days before the conference, and there were 2100 individuals who participated in the digital consultation that we ran through Mersal on May 31st.

It is important to note Tagadom’s role in garnering this level of engagement and participation in the consultations before and during the convention — Tagadom membership not only includes political parties, but also civic and revolutionary actors who have a high level of trust among the Sudanese public. This in itself points to the importance of having diverse representation in formal processes, and shows us just how much the public desire accessible, inclusive platforms to participate in these very spaces.

Local ownership

After the Tagadom convention, Mersal became its own entity. It continued to run consultations, serving as one way to fill an evident need amongst the Sudanese public to keep a safe space for dialogue running. The results and outcomes of the consultations are still shared regularly with the different Sudanese political actors, civil society, and mediators, but also with international organisations and mediation actors such as the UN Office of the Personal Envoy for Sudan and EU Office of the Special Envoy for Sudan.

Build Up is still responsible for finding ways for Mersal to be funded and operate technically. However, it was created as a tool for democratic processes and local conversations for the Sudanese public, and ensuring that it remains Sudanese-governed and owned is a crucial priority and responsibility. For this purpose, since June 2025, Mersal has been strategically guided by a steering committee of diverse Sudanese actors and activists. It is this committee that decides on the strategic direction of the work, and that chooses and approves which questions to prioritise and launch in the consultations.

Mersal was created as a tool for democratic processes and local conversations for the Sudanese public, and ensuring that it remains Sudanese-governed and owned is a crucial priority and responsibility.

The governance system of Mersal is explicitly written in a ToR document, which, alongside the names of the people in the steering committee, is publicly available for full transparency. This transparency and the model of ownership that has been established throughout the process is a key aspect of Mersal’s methodology and an important factor in the results it is generating. It has helped build trust with the audience, and has kept it alive and running throughout these two years.

What else made it work?

What made Mersal work was first and foremost its accessibility and ease of use. By using a WhatsApp, we were able to reach users through a platform that they are already active on. Moreover, the chatbot flow functions through a decision tree, giving brief, bite-sized explanations about the platform. This includes all the information necessary to understand and use Mersal, from the governance of the platform, to the consultations methodology, to the social media links to Mersal. The user is always just two clicks away from accessing the digital consultations, and once they click on the deliberation platform link, everything is on one screen: the question, the space to submit statements, and the voting options.

Polis, the platform Mersal deliberations operate on as mentioned above, is a key part of Mersal’s accessibility. Polis is an open-source digital deliberation platform developed by Computational Democracy Project. It enables large groups of people to share opinions in their own words while using statistical analysis and machine learning to identify patterns, areas of consensus, and points of disagreement. The platform has been used globally by governments, researchers, independent media organisations, and civic groups. Simply put, through Polis you can create an easy and straightforward interface for dialogue. A question appears at the top of the screen, and users can submit answers through a text box. At the same time, they can see what others have written and can vote on their statements by either agreeing, disagreeing, or passing.

Beyond its simple and seamless interface and the real-time participation of a large number of users, what makes Polis successful for democratic engagement on a platform like Mersal is the way its backend operates. For Mersal, the insights it offers facilitators in real time include an analysis of community statements, identifying which statements are divisive, which reflect majority opinion, and which demonstrate consensus across different opinion groups. An opinion group is made up of participants who tend to vote similarly across multiple statements and distinctly from other groups. This helps in understanding people’s positions across different issues and the linkages between them. Rather than simply prioritising what the largest group agrees on, Polis highlights areas of agreement across different groups, including smaller or dissenting ones. By examining results through the model of opinion groups rather than majority opinion alone, Polis’ analysis helps reduce the risk of the tyranny of the majority.

Rather than simply prioritising what the largest group agrees on, Polis highlights areas of agreement across different groups, including smaller or dissenting ones.

Another important element in Mersal’s wide use is its anonymity and confidentiality. People are more likely to participate in politicised conversations when their identities are hidden, especially in contexts where anonymity becomes not only a question of ease but also of safety. In the context of war, especially one where all populations are impacted, many communities may be afraid of scrutiny for sharing their political opinions. Moreover, as surveillance technologies grow and become more integrated in our daily lives online, it is essential to create grassroots platforms that prioritise privacy as a political act.

Limitations

While the anonymity and self-selection of the participants are core to the accessibility and democratic process that Mersal aims to facilitate, the lack of demographic information also means that the data we receive is not necessarily representative. It is subjective to the opinions of those we are able to reach and those who feel comfortable enough or are encouraged to participate. This is not necessarily something to be changed — collecting demographic data, even without names, may shift usership and user dynamics in unwanted ways. However, it is most definitely something to keep in mind when we discuss results, and perhaps more importantly, when actors from formal processes such as conferences want to include public voices. In such a scenario, Mersal, like any other public platform, is not a stand-alone representation of the Sudanese population, but should rather be understood as one part of a holistic approach to consultations.

Mersal is not a stand-alone representation of the Sudanese population, but should rather be understood as one part of a holistic approach to consultations.

One risk that emerges from anonymity is the possibility of coordinated attempts to game the system. We have encountered this in two forms. One is through flooding the platform with hate speech or completely irrelevant statements. To tackle this, a criteria for accepting and rejecting submissions is an important element in creating a safe and democratic space for dialogue. Mersal now filters out hate speech or statements deemed irrelevant, and these criteria were communicated transparently to participants through the information provided in the chatbot flow. A more difficult challenge is when a group coordinates voting behaviour in an attempt to skew the results. However, because of the way Polis operates and analyses participation patterns, coordinated behaviour tends to appear as its own distinct opinion group within the analysis, since these participants vote in a similarly coordinated manner across multiple statements.

Connectivity of course is also a challenging issue. When people do not have access to the internet or a device, they simply cannot participate in an internet-based platform. Moreover, while Mersal is made for and reaches Sudanese people living all around the world, those living in Sudan — those most impacted by the war — are also more likely to experience internet black-outs or electricity shortages, either as a result of socio-economic crises or as a tool of the war itself.

Accessibility more broadly always has its limitations as well. Here, we need to think about digital literacy, awareness of the platform, and people’s connectivity to networks and information through social media or other mediums.

Finally, sustainability is always something to think about. On the one hand, as a participant-centred initiative Mersal does have the potential to remain active for as long as users would like it to be. However, financial and human resources as well as technical know-how are necessary for Mersal’s growth and continuation. We hope that the model that we have developed with the steering committee will eventually lead to ways that these needs can be fulfilled and run completely by community members themselves.

Future of Mersal

As more and more people join Mersal, it continues to grow and expand its community with it. It also continues to gain more recognition and traction as a deliberation platform by local and international actors. But one question remains: what benefit do consultations and conversations around political issues serve as long as the conflict parties are not open to negotiations and the political process is in a deadlock? The answer to that from my perspective is that democracy is built through active participation. It is therefore essential for Mersal, and any other deliberation platform no matter the context they were created in, to be people-centric and process-oriented, to respond to user feedback, and to remain accessible and engaging. These spaces and the actors who want to access them must never forget that the main audience is the people themselves, as they are the force that drives democracy.

Not every peacebuilding challenge is a data challenge, but all data choices are peacebuilding choices. In conflict environments, what information gets collected and from whom, what categories or labels are chosen, and who benefits from the insights are technical questions with power-laden answers. Artificial intelligence (AI) tools are increasingly available to help peacebuilding practitioners face data challenges, such as analyzing media coverage across languages, conducting public consultations, or understanding how divisive narratives spread. Some of these tools can help increase inclusion in mediation and peace processes. Others simply make more operational tasks easier and faster. The utility of many more is being tested with open fields for experimentation and learning. Whether AI tools advance peacebuilding depends less on the sophistication of the data or its handling than on the mindful choices practitioners make about its use.

AI tools are most useful when they are embedded within and supporting people-centered peacebuilding processes, not the other way around. Responsible AI discourse rightly emphasizes keeping “humans in the loop,” meaning that human input and judgement is integrated at key stages of a model’s workflow. An inversion of this principle, putting AI in the loop, is an invitation for peacebuilding practitioners to intentionally use AI (or not) as a constrained support tool within the processes, practices, and values known to be effective and ethical.

De-hype and Dig In

The effective use of AI in peacebuilding starts with clarity about needs over novelty, and specificity about how AI addresses them. For example

• We need to understand how online discourse is entrenching and escalating the conflict. We are using supervised text classification to analyze content inciting violence on social media.
• We need to monitor compliance with ceasefire commitments. We are using computer vision to flag potential violations in satellite imagery for human review.
• We need to make sense of a wide public input process. We are using pattern-detection tools to organize open-ended consultation responses so facilitators can engage with dominant concerns while still identifying outlier and minority views.

When we move beyond broad discourse to de-hype AI with precision and dig into the real-world applications involved, we make our assumptions and limits visible for those we’re working with to meaningfully assess value and risks. At the same time, many assumptions and limits only become visible through practice and experimentation. Digging in means tolerating initial uncertainty, often starting small and building with, instead of for, people.

With Participation

Peacebuilding practice centers on participation, partnership, and ‘right relations.’ This includes engaging with affected communities in decisions about the collection, interpretation, and use of data. When AI tools promise efficiency, scale, or synthetic aggregation, there are potentially high costs involved for the participatory process. Many uses of AI can turn participation into a proxy rather than a practice by replacing engagement with summaries, predictions, and synthetic representations.

It is technically possible, for example, to build a hate speech classification system trained on synthetically-generated data, governed by labeling rules created without community input, and validated primarily by large language models rather than affected groups. Look at how many ‘people problems’ can be solved by removing people!

Consider an alternative example from Build Up, where Sudanese practitioners engaged in deep dialogue as they worked together to train a custom hate speech detection classifier based on social media content evolving through the current conflict in real time. In a similar process in Kenya, looking instead at what counted as ‘polarizing’ content, partners moved from skepticism about contextual AI limitations to ownership over what they called “localized” tools. One summarized: “We can build a tool that works for us.”

To Shape AI Use

Technocracy promises that complex social problems yield to better data, smarter models, and more efficient systems. Peace policy and practice suggest a different story for problem-solving. Dialogue processes work when participants feel heard, not accurately summarized in a generated report. Early warning systems fail not for lack of data, but because the political will to act on warnings is absent. These are not problems AI solves, and AI should not become the latest bolt in a technocratic gatekeeping process that deprioritizes people doing the slow, relational work that transforms conflict.

However, resisting technocracy should not mean foregoing technology. Global peace practitioners should not shy away from engaging and experimenting with AI tools. In my experience and from participation in a community of digital peacebuilding practice, practitioners know what they need to reach more people, to expend fewer resources, or to move a peace process forward. The key is to put AI in the loop of peacebuilding processes, where practitioners are not just end users of technology developed elsewhere or become beholden to AI-powered analysis. Instead, practitioners should be trained, trusted, and resourced to lead in actively defining and shaping AI tools that serve current and emerging peacebuilding needs.

Originally published at Peace Policy: Solutions to Violent Conflict, №62 (2026), Kroc Institute for International Peace Studies. You can download the PDF of the full issue here.

Julie Hawke is the Peacebuilding and Research Lead at Build Up, and is also pursuing a joint Ph.D. in peace studies and sociology at the University of Notre Dame through the Keough School’s Kroc Institute for International Peace Studies.

This blogpost is part 3 of a 3-part series on the Polarization Footprint, a metric of affective polarization on social media platforms. Part 1 explores why we think it is important to measure affective polarization; part 2 shares the results of the polarization footprint in Kenya; part 3 explains the methodology for arriving at this metric.

The polarization footprint is a cross-platform measure that ranks social media platforms according to the prevalence of affective polarization present on the platform. The polarization footprint relies on observing content and relationships on social media platforms, and is designed in such a way as to be comparable across platforms. The methodology is explicitly designed to deliver defendable and confident measures of the minimum prevalence of polarization.

To complement the polarization footprint, we suggests running the Neely Social Media Index survey alongside this on-platform observational measure. The objective of the survey is to explore positive and negative user experiences on social media platforms. By running the survey together with the on-platform measure, we can derive some meaning from the relationship between their results, and understand the differences between perceived negative experiences and observable polarization.

The full methodology as it was used by Build Up in Kenya in 2025 is available here. What follows is a descriptive summary of the methodology.

Three types of affective polarization

In order to capture the prevalence of affective polarization we have developed a composite measure made up of three parts, each measuring a component of affective polarization — attitude polarization, norm polarization and interaction polarization. These three components are measured separately to isolate specific aspects of the dynamic.

• Attitude polarization is characterized by perceptual shifts and often expressed as fixed and overgeneralized beliefs or notions about certain groups, identities, and their intersections. We measure attitude polarization by examining whether the language used in social media posts seen by a representative sample of users within a community contains negative stereotypes, dehumanization, deindividuation, vilification or a call to violence.
• Norm polarization is at play when combative norms of interaction reinforce the erosion of trust between people. We measure norm polarization by identifying the prevalence of challenges, or lack thereof, to polarized attitudes in the comments in a social media thread below a post, when either the post or one comment expresses attitude polarization (where a low prevalence of challenge signals a norm shift towards negative expectations).
• Interaction polarization results in a fragmentation of narratives, user experience and truths that ultimately impacts both the interests and affiliations of people, creating a self-reinforcing cycle of polarization important to social cohesion. Whereas we measure attitude and norm polarization based on individual behaviors (what content gets posted, what reactions items of content get), interaction polarization is a network-wide dynamic, not a post-level attribute. It is the extent to which users of a social media platform fragment into dissimilar clusters viewing and interacting with content in different universes. We measure interaction polarization as the degree of fragmentation in the hypergraph showing extent of overlapping content seen and accounts followed by participants.

Methodology summary

We gather data by running online ads to solicit the participation of a representative sample of social media platform users for a given region or area. Once participants have been selected they download a custom app, through which they are prompted to log in to the website of the social media platform on which they were recruited. With the participants’ consent, the in-app browser then scrapes the most recent posts and comments from their feeds. Local peacebuilding experts then manually annotate any content that includes attitude polarization within a representative subsample of the collected content from each platform. With the consensus labels (that take into account where annotators agree or disagree), we arrive at a measure of the prevalence of attitude polarization within the platform’s entire dataset (in a particular context), including a confidence interval for this measure.

We also use the annotated dataset to train a text classifier that can identify the majority of attitude polarizing threads in the entire dataset. A subsample of threads containing attitude polarization are then annotated by the same experts to identify instances of “challenge” to that, thus arriving at a measure of the prevalence of norm polarization for each platform.

Finally, we construct the hypergraph, with one hyperedge for each account which was followed or from which a post was seen in a feed. Participants share hyperedges for each account that they both saw or followed (if any). We then quantify sorting as the total correlation between the indicator random variables that correspond to whether a randomly chosen participant belonged to a given hyperedge, and this is the interaction polarization score at the platform level.

Each of these measures results in a numerical rank for the platform; adding them together gives us the polarization footprint for that platform.

To complement the on-platform data analysis that we undertake on the social media feed content viewed by participants we also carry out a survey asking the same participants their positive and negative user experiences on social media platforms. By running the survey together with the on-platform measures, we can derive some meaning from the relationship between their results, and understand the differences between self-reported negative experiences and externally observable polarization.

A robust, ethical methodology

The polarization footprint thus offers an evidence based metric against which both public authorities and individual citizens can make comparative assessments of the polarization leve on different online platforms. The methodology is not only robust, but is also guided by a set of ethical principles:

• Data protection & privacy — We obtain informed consent from participants; only one researcher can access information known to be personally identifying (known PII); only team members see the text of the posts (including non-public ones) for annotation and classification; we only publish aggregated results; all social media data is deleted at the end of the study and all PII from the survey is deleted at the end of the study; people can withdraw their participation or request the deletion of their data at any time.
• Review — Before commencing any data collection, the study undergoes research ethics review at an accredited academic or research institution in the country.
• Pilot — Before committing to a adapted methodology for a specific context, we conduct a pilot (a scaled down version of the full study with fewer participants) and make any required changes.
• Pre-registration — After the pilot and before commencing data collection for the main study, the study design and data analysis methods are pre-registered on OSF.
• Transparency — All results are made publicly available in a report and on a website.
• Open source — The code for producing the annotation servers, IRR metrics, prevalence metrics and classifier model are available here; the attitude polarization model is open source and made available on Hugging Face.

Got questions or comments?

We’re keen to hear feedback on this methodology, and to make it accessible for others who want to use it. Reach out to us on team@howtobuildup.org and remember you can download the full methodology here.

This blogpost is part 2 of a 3-part series on the Polarization Footprint, a metric of affective polarization on social media platforms. Part 1 explores why we think it is important to measure affective polarization; part 2 shares the results of the polarization footprint in Kenya; part 3 explains the methodology for arriving at this metric.

The Kenya Polarization Footprint measures how social media is dividing Kenyans. Specifically, we measure “affective polarization”, where people dislike and distance themselves from others based on their identity. This is not the same as a difference of opinion: differences of opinion are good for democratic debate, and for our diverse society. Identity-based polarization is not: increasing dislike, distrust, and animosity towards other social groups results in societal divides that can eventually lead to violence.

The way content is shared on social media makes societal divides wider. Measuring just how much online polarization is present on each social media platform is an important first step to defending our society. For more on why we measure affective polarization, have a read of part 1 of this blogpost series. The full results report is available here, or just read on for our five main findings.

#1: X (Twitter) is the most polarizing platform in Kenya

The polarization footprint is a composite measure made up of three parts, each measuring a component of affective polarization — attitude polarization, norm polarization and interaction polarization. Across all measures of polarization that impact societal divides, X (Twitter) does worst. This means that when you scroll through content on X, you are most likely to experience polarization.

We rank platforms for each type of polarization, and then combine these rankings into a single overall ranking, which is what is shown in the table above. The three components are important to think about individually:

• Attitude polarization looks at how much language used in social media posts and comments expresses negativity towards an identity group, which we call “othering” (more othering = more polarization);
• Norm polarization looks at how often people challenge polarized attitudes in the comments in a social media thread (no challenge = polarization is “normal”);
• Interaction polarization looks at how much users of a social media platform are fragmented into different content universes (more fragmentation = more polarization).

#2: In Kenya, social media narratives on X and YouTube are very fragmented

There is one aspect of affective polarization that is particularly striking: we are all experiencing different versions of it. Interaction polarization is the extent to which users of a social media platform are fragmented into different content universes, which impacts both the interests and affiliations of people, creating a self-reinforcing cycle of polarization.

In the table above, the higher the measure, the more fragmentation of narratives, experiences and truth in our society. The scores above tell us that on X and YouTube people live in their own social media content universes separate from the universes of others.

These scores are also important to the overall ranking of platforms: Facebook and TikTok have a middling ranking for attitude and norm polarization, but on Facebook the interaction polarization score is much lower, meaning that people’s content networks are less fragmented. That’s what ends up making Facebook the least toxic platform, with TikTok ranking third most toxic.

#3: An average Kenyan spends up to 21 minutes per day reading threads that express negativity towards an identity group

Within these fragmented universes, Kenyans are seeing content that expresses negativity towards an identity group daily. The table belows details the minimum % of attitude polarization in the average person’s newsfeed. What this doesn’t tell us is how long people spend looking at each of these polarizing threads, but it seems reasonable to hypothesise that they hold our attention. Even more, this is a very conservative measure: 5 or more experts had to agree a post or comment was polarizing for us to count it.

To understand what these results mean for a normal social media user in Kenya, here’s one example. We expect a minimum of 3.81% of threads on X (Twitter) are polarizing. This means that for every hour you spend on X, on average you’re spending a minimum of just over 2 minutes on polarizing content. 55% of Kenyans are spending 3–9 hours on social media per day; if that time were spent on X, then your average Kenyan would spend at least 7 to 21 minutes exposed to polarizing content. Imagine being shouted at for 7 to 21 minutes every day. No wonder we feel angry.

Imagine breathing from a car exhaust pipe for 7 to 21 minutes every day. Polarizing content is making us sick.

What’s worse, some Kenyans (5%) spend more than 12 hours on social media per day, which means they are breathing in polarization for a minimum of 27 minutes every day; getting shouted at for a minimum of 27 minutes every day. These super users are often also the most prolific on social media — the influencers, the rabble rousers, the ones who drive narrative. The more they see, the angrier they get, the more toxicity that goes back out into our society. It’s a vicious cycle.

#4: Most of the time, Kenyans don’t challenge polarizing language on social media

As with the attitude polarization scores, our measure of norm polarization is conservative and reports the minimum % of polarizing threads where no-one challenges the polarization.

Imagine that you are being shouted at, and no by-stander steps in to defend you or to lower the tone. That’s what norm polarization is about.

When it comes to norm polarization, while X still has the worst score, the other platforms are close behind — meaning there are few challenges to polarized attitudes when they are expressed in posts or comments on all platforms. No challenge signals a norm shift towards negative expectations — that is, it signals that attacks on others are considered normal on this platform.

#5: On Facebook and X (Twitter), almost 50% of Kenyans report seeing content that is “bad for the world” in the past 4 weeks

Observing what polarizing language, challenges to this language, and patterns of interaction look like on social media platforms is an important measure of how these platforms are affecting Kenyan society. But we worried that we wouldn’t get the full picture of the impact this is having on Kenyans unless we asked directly about their experiences on social media.

We asked 5000 Kenyans whether, in the past 4 weeks, they had negative experiences on a social media platform or saw content they thought was “bad for the world”. On all platforms except YouTube, over 30% of the people we surveyed had a direct negative experience. The same was true of content they considered to be “bad for the world”.

These results tell us one thing clearly: X (Twitter) is a toxic platform, across all our measures. They also leave us with some questions: what is happening on Facebook and YouTube for people to self-report so differently to what we observe in their feeds? We’ll be taking that question forward with us, and conducting some further analysis of the survey results to try to unpack what is going on.

What next for the polarization footprint in Kenya?

The polarization footprint in Kenya has been developed in partnership with the National Coalition on Freedom of Expression and Content Moderation in Kenya (FECoMo), of which we are members. Over the coming weeks and months, we will be discussing these results with FECoMo members, and figuring out how we might take them forward.

We set out to measure affective polarization on social media in Kenya because we are worried that disagreements online spiral out of control, and that this happens because they become about people rather than ideas. Or in the words of Rachel Olpengs, Lead Coordinator of FECoMo: “The Kenya Polarization Footprint reveals deepening societal divides driven by affective polarization on social media where identity-based hostility fuels distrust and animosity. This growing digital harm is largely enabled by the presence of super users who perpetuate a cycle of division and platforms receiving perverse incentives from polarized content further fragmenting public discourse and posing a tangible threat to Kenya’s social fabric. The report underscores that polarization is not merely a digital nuisance but a measurable social harm.”

“ Addressing [the polarization footprint] requires a shift in perspective from individual responsibility to collective action. It demands platform accountability, participatory governance, and systemic interventions to safeguard democratic discourse and social cohesion.” — Rachel Olpengs, Lead Coordinator, National Coalition on Freedom of Expression and Content Moderation in Kenya (FECoMo)

The polarization footprint gives us an anchor from which to develop policies and campaigns that can help Kenyans take back their digital space. Join us as we take this forward.

→ For details of how exactly we conducted this measurement, head over to part 3 of this blogpost series.

This blogpost is part 1 of a 3-part series on the Polarization Footprint, a metric of affective polarization on social media platforms. Part 1 explores why we think it is important to measure affective polarization; part 2 shares the results of the polarization footprint in Kenya; part 3 explains the methodology for arriving at this metric.

All is not well in our digital ecosystem. The social media platforms through which we engage with what is going on in the world are designed to optimise for engagement. This business model underpinning how news and public debate gets prioritised tends to surface inflammatory and divisive content and risks fragmenting users into separate information bubbles. The voices and perspectives that are recommended online are not representative of the spectrum of public opinion but skewed towards polarized, extreme or contrarian positions. Our cognitive bias toward sensationalist content interacts with recommender algorithms to create feedback loops where we see more and more extreme content.

Online polarization has real world offline consequences. It affects which topics are discussed and how we discuss them in our families, workplaces and communities. This shaping of civic discourse frames political debate with consequent impact on electoral outcomes. Social cohesion is jeopardised by divisive narratives that get amplified triggering spirals of threats and violence against ‘out groups’ within our society or channelled toward foreign actors.

But can we actually measure where, when and how online polarization is happening, and on which specific platforms? And then what can be done about it?

We have developed a conceptual framework called the polarization footprint to answer this question, and a methodology to calculate this polarization footprint across social media platforms.

What is the polarization footprint?

The polarization footprint is a cross platform measure that enables comparing and ranking social media platforms on the prevalence of affective polarization. We define affective polarization as relationship or identity-based (not opinion or idea-based), where increasing dislike, distrust, and animosity towards other social groups results in societal cleavages becoming salient. Affective polarization is a dynamic psychological process intertwined with conflict escalation, by which a self-reinforcing spiral operates to separate ideological or identity based groups into increasingly distanced and aggregated adversaries. It is relationship- or identity-based in that its focus is the increasing dislike, distrust, and animosity towards those from other parties or groups.

This differs from issue-based polarization, which focuses on the disagreements or ideological distance between individuals on policy areas of the type that we are familiar with in opinion polling on public policy issues. Just as some conflict can in certain cases be constructive, issue-based polarization is not necessarily problematic. In contrast, affective polarization can increase the risk of escalation to violence by taking a conflict that is more specific and localized toward something more general, identity-based and antagonistic. Issue-based polarization becomes affective and intractable when we cannot change what we think or say without losing core social networks or identities.

Our proposition is that affective polarization is a systemic risk in the digital ecosystem that should be monitored and that may require on-platform mitigation measures.

How does affective polarization impact us?

A core maxim of peace mediation is to “separate the person from the problem”, that is to encourage parties to address the content of the message rather than focusing animosity on a group identity of the person expressing the message. But, as mediators can attest, it is not easy to assist parties to make this shift and to escape the affective polarization trap. When affective polarization is wide-spread across groups in society, conflict can become intractable: mental categories become rigid and de-escalation becomes very difficult.

We also think that affective polarization is one of the reasons for limited success of fact checking and debunking of online disinformation and hate speech in the absence of a wider set of interventions. Those who are psychologically invested in a group identity are not primarily concerned with what is true or false but rather with the emotional ‘in group’ bonds.

We believe that affective polarization, measured by the polarization footprint, poses a systemic risk to our societies because because it identifies direct attacks on human dignity and inhibitions to freedom of expression (with consequent impact on freedom and pluralism of the media), points to deteriorating civic discourse and heightened risk of public tensions, and can ultimately offer a signal for risk of violent conflict and ensuing dramatic consequences for physical and mental wellbeing.

The problem of polarization has repercussions across policy domains. There has been much attention to whether and how polarization has skewed public debate on issues such as climate change, vaccine uptake, safety of minors, gender politics and electoral integrity. As a peacebuilding organisation, Build Up’s specific concern is the impact of online polarization on offline risks of violent conflict, possibilities to reach peaceful resolutions during war and the ability to sustain peace processes.

The polarization footprint is our answer for how to measure the systemic risks online platforms pose to peace, and how conflict mitigation can be incorporated in our approach to technology governance.

So what can be done about the polarization footprint?

Research on digital platforms shows that content with affective polarization does something else: it captures attention online, it increases engagement. That’s why social media companies that rely on capturing attention as a way to make money have no incentive to control for polarization. We argue that polarization should not only be considered a digital harm that drives conflict, but also a negative externality of the attention economy. That’s why we call our measure the polarization footprint — it is to the attention economy what the carbon footprint is to the industrial economy.

The polarization footprint is to the attention economy what the carbon footprint is to the industrial economy.

The polarization footprint offers a compelling new lens through which to analyze and address this negative externality. We have developed a methodology to measure polarization footprints. Once the risk is measured it needs to be mitigated. Integrating the polarization footprint into existing and future policy frameworks holds significant potential. We are begining to explore, with partners, what this might look like. Some initial ideas emerging:

• One opton is to price polarization into production costs through taxation as a means to change the business model of technology companies. Here is more on that idea.
• The polarization footprint could provide an evidence base for regulators tasked to enforce the European Union’s Digital Services Act (DSA). We’ve explored this idea in this report.
• We’re thinking through how the polarization footprint could be used to inform consumer choice. This can be a powerful alternative tool to remedy negative externalities. We’re looking at examples from other policy areas where this has been done, such as the nutri-score for food labelling, the blood diamond campaign and the eco label for environmental impact, all of which provide easily understandable information to consumers allowing them to choose between similar products.

Over the coming months, we will continue discussing the policy implications of the polarization footprint — we see this as a contribution to the global debate surrounding mitigation of systemic risks to peace and security stemming from digital technologies.

→ Next up: read the results of the polarization footprint in Kenya.

→ Skip ahead: read the methodology to measure polarization footprints.

How do social crises shape the way we live together? This was the starting question for Zwischentöne — Learning from Crises, an initiative we ran in Lucerne, Switzerland.

In international peacebuilding, telling life stories is regarded as a resource for creating meaning, fostering empathy, and helping both individuals and societies to process traumatic experiences (Bar-On, 2006; Lederach, 2005). Storytelling can build bridges and open up a space for dialogue. This is why we opted for a biographically inspired approach, leaning on oral history, in which 70 people documented their experiences of societal crisis.

The five insights that follow are offered as encouragement for anyone looking to enhance conflict management and build greater crisis resilience in Switzerland and beyond.

1. Appreciation of experiences

“Today, everyone in our family is trying hard, but as a mother, I notice that something is still brewing, and that a lot of unspoken and stressful things are still resonating. And in general, I find that many people, even before the coronavirus, are very oversensitive to everything.”

A woman from Rothenburg

In times of social crisis, we often only perceive the loudest voices. We lack a deeper understanding of how people actually live through and cope with upheaval. Particularly the experiences of marginalised groups — such as those affected by poverty, young people, refugees, or people with disabilities — disappear in pale statistics. People are talked about rather than talked with.

Zwischentöne (“nuances”) aims to honor individual experiences with crises and conflicts in all their complexity and uniqueness. People from the canton of Lucerne documented their experiences in open group meetings, or submitted them in writing or via voice message. These voices are made publicly accessible on zwischen-toene.ch, in the LU222 archive of the Lucerne Museum, and through media contributions. This creates a contemporary historical documentation of our crisis-ridden present — beyond the fleeting moment of a story shared orally.

Our experience shows that dialogue and listening formats are only effective when participants feel a genuine, empathetic interest. This above all requires a credible and consistent inner attitude, and a willingness to engage again and again with openness and curiosity towards different lived realities. At the same time, this means that dialogue formats must first provide space for subjective experiences, before inviting people into conversation about wider social tensions and experiences of crisis.

2. Deep need to be understood

“Many Swiss think: refugees come here, they get money, they get everything, and they do nothing. But that’s not true. We always try everything: How can we integrate? How can we find a good job?”

Somi from Lucerne

For many contributors, it was deeply meaningful to speak openly about their own experiences and to be listened to by others. Several participants shared stories they had never voiced before, along with the sense of not having been understood, or of being misunderstood in the past. Such repeated experiences of being disregarded weaken democracy: those who feel unheard gradually withdraw. The longer this sense of exclusion persists, the less empowered people feel to take part in existing opportunities for dialogue and participation.

This withdrawal is reinforced by a psychological factor that is often overlooked: invisibility. Many people with refugee backgrounds, physical disabilities, or living in precarious circumstances experience — often on a daily basis — that their needs are not taken seriously by mainstream society. At the same time, many dialogue and participation opportunities themselves appear inaccessible. This fosters a form of self-disempowerment,captured in sentiments that participants voiced, such as: “I didn’t know I was even allowed to take part.”

The depth of this need to be heard was also reflected in people’s response to our more experimental formats of story collection. Some readily agreed to record their experiences of crisis while sitting on a park bench during a Sunday stroll. This willingness surprised us in what is often considered a reserved society. Perhaps it was precisely the relative anonymity to speak to a stranger that made people want to share.

3. Meet people where they are

“It also bothers me a lot that I feel so lonely so often, especially in Switzerland. It has something to do with the current society and perhaps also with my retreat into the silent corner. And most likely, there’s just a general loss of trust that has happened over the years.”

I. from Lucerne

Our experiences with Zwischentöne highlight how important it is to meet people in the places where they spend their everyday lives and feel at ease. We were invited into libraries, parishes, and intercultural meeting points. Choosing these locations deliberately signalled our respect for structures that already foster exchange. This approach was only possible thanks to the many organisations that were willing to collaborate. The broad strategic network included actors such as Caritas, Sans-Papiers-Beratung, the Lucerne Red Cross, the Kriens Library, and many others. Through them, we were able to reach a wide variety of groups. Acknowledging and valuing existing structures also increased our acceptance as outsiders.

Reaching people where they are also means going beyond urban centres. Cities are often well served by participation formats, while rural areas tend to be overlooked. An interesting exception is the MiPaPe association, which organized an exchange in Glarus in September 2025 — a place few dialogue initiatives look towards. Engaging rural regions with cantonal or national projects from outside requires time to build relationships — but the effort is particularly worthwhile.

One group we struggled to reach were young people, despite the involvement of dedicated youth organisations within the network. We also sought dialogue through Instagram. While a few meaningful conversations developed there, only a small number accepted the invitation to document their experiences. In retrospect, this was likely not only due to our communication, but also to the absence of young people within the project team — people who could have acted as credible mediators and bridge-builders.

4. Linking personal experience to political processes

“Our authorities determined the measures with this blind faith in science. That wasn’t good. At the same time, it was, of course, difficult. I wouldn’t have wanted to be Federal Councilor Berset back then. He certainly had a difficult job. But for a revolutionary like me, the understanding of democracy was simply lacking.”

Man from Rothenburg.

The Zwischentöne experience shows how much the artificial separation between individual experience and political practice can intensify social tensions — and that bridging this gap is everyone’s responsibility.

Those who hold political or administrative power also have their own experiences of crisis. They are part of society, with doubts and stories of their own, and under pressure when crises unfold. Recognising the human face of politics can help to dismantle barriers. As a society, we could learn to deal with crises in more resilient ways if those in positions of political responsibility were willing — and able — to show their vulnerability not only in confidential, nuanced conversations, but also publicly, especially in times of crisis. At the same time, we all need to reflect on the tendency to level sweeping accusations at “politics.” Society as a whole must hold up a mirror to itself and consider which opportunities for participation it is not making use of. More self-efficacy and empowerment would foster greater courage and hope.

Civil society, too, must reconsider its relationship with politics. It is not enough to simply point out grievances; we must also understand how decisions are made and where we can make a tangible contribution. Zwischentöne has shown that the doors are much more open than many imagine: our exchange with parliamentarians in the Federal Palace in March 2024 was direct and on equal terms. At our big public event on 16 September 2025, we brought decision-makers into dialogue with the Zwischentöne narrators. Many agreed spontaneously — some even without knowing the project in detail. This shows that civil society can play an important role in narrowing the distance to politics.

5. Build alliances across forms of discrimination

“There’s a well-known saying: “You can judge the strength of a society by how it treats its marginalised groups.” I would add: “… in times of crisis.” In ordinary times, we may not be wanted, but at least we are accepted. In crises, however, we become invisible. Before the next major crisis, we need to learn that only a truly diverse society can be strong. Everyone must have a seat at the table if we are to find solutions together.”

Jahn Graf

Zwischentöne set out to reflect a cross-section of society. We succeeded: stories from people with very different backgrounds and opinions now stand side by side on the website and were discussed together at the public event. Our particular focus was on reaching marginalised groups and fostering connections between them. The guiding idea was simple: if we consider different forms of discrimination together, we can raise greater political and social awareness. The minority becomes the majority.

The perspectives of young people, those experiencing poverty, people with disabilities, and those with refugee backgrounds are often treated in isolation. Many of these groups have had to — and still have to — fight hard to make space for their concerns. As a result, many communities organise separately, even though they share similar experiences of exclusion.

This often leaves little space for a nuanced understanding of overlapping forms of discrimination. As a society, we could make far better use of this potential — for deeper mutual understanding and for building new alliances across identities.

And now?

Building on the insights gained from these stories, Zwischentöne is now shifting its focus towards the institutional level. In cooperation with public authorities, we are developing a prototype for participation in crisis management at both municipal and cantonal levels. In this way, we are helping to make the “politics of being heard” an even stronger reality.