Artificial Intelligence (AI) for Peace.

AI is not only being used to shorten kill chains and enhance coordination capacity in wartime. A parallel set of applications, smaller in scale but expanding quickly, uses the same underlying technologies to support peacebuilding, mediation, atrocity documentation, language access, and early warning of conflict. A September 2025 meta-review of more than 600 documents in the field found that these tools are showing consistent value across multiple use cases, but face acute and persistent gaps in funding, evaluation methodology, and human-rights safeguards. In budget terms, documented AI for Peace programs operate three to four orders of magnitude below the scale of frontier-AI investment, as shown in Figure 5.7. The asymmetry is so large that AI for Peace is best understood not as a counterweight to military AI, but as a parallel ecosystem that has emerged in spite of, rather than because of, the dominant investment pattern.

The most developed applications of AI for Peace are in mediation and inclusive dialogue. Within the UN system, the Innovation Cell of the Department of Political and PeacebuildingAffairs has built machine translation, sentiment analysis, and large-scale text-clustering tools into active mediation work across Libya, Yemen, Iraq, Lebanon, Haiti, and Bolivia. These tools allow mediators to process community submissions at a scale that would otherwise be impossible to read manually, and they make it harder for any single faction to dominate the consultation record. Outside the United Nations, civil-society mediators have run similar processes. In 2023, the Crisis Management Initiative used the Remesh platform to run a real-time Sudanese women’s dialogue with up to 1,000 participants, building on earlier processes in Yemen and Libya. The Centre for Humanitarian Dialogue’s Libyan National Conference Process, though mostly in person, brought together 7,000 participants across 75 sessions in 43 locations and reached an estimated 1.8 million Libyans through digital channels. In November 2024, Conflict Dynamics International launched Akord AI, the first AI co-pilot built specifically for peacebuilders, trained on around 1,500 documents on the Sudan conflict and operating in Arabic and English. AI is also moving into structured deliberation at a scale that human mediation alone cannot match. A 2024 study tested an AI consensus-building system known as the Habermas Machine with more than 5,700 UK participants on divisive issues including Brexit and immigration. The system produced consensus statements that participants rated clearer, more informative, and less biased than those produced by human mediators, and it measurably reduced group division. In parallel, the open-source deliberation platform Polis has been used at national scale through Taiwan’s vTaiwan process, where roughly 80 per cent of the 26 issues debated led to government action between 2015 and 2018, making it one of the few examples of structured public deliberation producing direct policy change at scale.

Language access is the second area where AI is making a measurable difference for peace. The free flow of information depends on whether people can participate in public life in their own languages, and a large fraction of the world’s languages are still under-served by commercial translation systems. Meta’s No Language Left Behind system, released in 2024, supports 200 languages and improves average translation quality by 44 per cent over previous systems, with a deliberate focus on lowresource African and South Asian languages. Smaller initiatives are filling specific gaps: CLEAR Global’s Gamayun initiative is working to bring ten marginalisedlanguages online for humanitarian use, and Lelapa AI, founded in Johannesburg in 2022, has released the Vulavula API and the InkubaLM small language model covering five Africanlanguages spoken by around 364 million people. At the delivery end, the International Rescue Committee’s Signpost humanitarian information service now reaches more than 20 million people directly across nearly 30 countries and 25 languages, with social-media reach above 100 million. AI is also reshaping accountability for atrocities. Documentation organisations are using machine learning to process volumes of open-source evidence such as satellite imagery, social-media posts, and leaked databases that would once have required years of manual review. The Berlin-based archive Mnemonic, which runs the Syrian, Yemeni, Sudanese, and Ukrainian archives, has worked in the preservation of more than 15 million digital records of human-rightsviolations. The Yale’s Humanitarian Research Lab has produced more than 65 reports during the El Fasher siege in Sudan, including a joint March 2026 report with NASA Harvest showing that Rapid Support Forces deliberately razed at least 41 farming communities, of which 28 were depopulated, with farmland cut by 82 per cent. The same lab’s Conflict Observatory verified 19,500 children deported from Ukraine to Russia and Belarus across at least 210 facilities. The same techniques are being used for post-conflict recovery. In demining, the HALO Trust’s partnership with AWS has appliedmachine-learning models to 11 terabytes of drone imagery from Ukrainian minefields. The result is a reduction in the time from data capture to actionable map from days to hours, which clears land for return and reconstruction much faster than was previously possible. In damage assessment, United Nations Satellite Centre (UNOSAT)’s October 2025 Gaza analysis identified 198,273 affected structures, of which 123,464 were destroyed, with roughly 81 per cent of all structures in the Gaza Strip damaged. Without automated satellite analysis at this scale, the magnitude of destruction would have become clear only months or years later, with significant implications for reconstruction planning, humanitarian access, and any future legal proceedings. These projects differ in maturity, geographic focus, and the quality of their evaluation, but they share a common logic: each uses AI to broaden participation, accelerate evidence collection, or shorten the feedback loop between communities and the institutions that affect them. They sit outside the dominant model of military AI procurement, with different funders, different incentives, and different metrics of success. In budget terms they remain a small fraction of total global AI investment. AI can already widen participation, strengthen peace processes, and help societies confront conflict without violence. Whether it does so in practice depends on resourcing and design, not capability. The capability exists; what is missing is investment at a scale that matches the technology’s potential, and a governance architecture that recognises peacebuilding applications as a public good, worthy of funding alongside rather than after the military and commercial uses that currently dominate the field.