[{"data":1,"prerenderedAt":269},["ShallowReactive",2],{"store-blog-data":3,"store-work-data":45,"menu-data":60,"footer-data":176,"insight-elevate-benchmark-and-technical-report-":177},[4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44],"elevate-benchmark-and-technical-report-","building-a-national-digital-media-literacy-strategy","what-canada-can-learn-from-polarization-in-the-us","can-ai-chatbots-safely-navigate-conversations-invo","confronting-palestine-israel-polarization-through-","listening-and-dialogue-for-countering-polarization","stopping-gender-based-violence-through-bystander-t","voices-a-new-model-for-digital-dialogue-on-israel-","testimony-on-civic-resilience-in-canada","canada-is-talking","transformational-conversations","bystander-blueprint","polarization-canada","polarization-in-the-public-square","try-facilitator","canada-needs-to-address-anti-asian-racism","innovative-solutions-hack-the-divide","how-does-ai-impact-trust-in-democracy","three-things-canada-can-do-to-address-transnationa","how-do-you-imagine-ai---try-aimagine","election-watch-foreign-influence","election-watch--pencils-ballot-boxes-and-poll-work","election-watch--canadas-election-avoided-violence-","election-watch-indophobia","how-to-protect-yourself-from-disinformation-transn","election-watch---carneys-election-as-liberal-leade","election-watch-canada-51st-state","understanding-opioid-and-substance-use-and-barrier","democracy-and-you-a-handbook-for-detecting-and-pre","evaluating-content-frameworks-resilience-to-imve","mapping-vulnerable-audiences-with-high-acceptance-","article-beijing-is-harassing-diasporas-in-canada--","community-consultations-on-foreign-interference-in","survey-from-the-foreign-interference-resilience-education-project","survey-from-the-public-safety-project-comet","when-to-play-games-during-a-pandemic","making-human-agency-the-default-in-the-connected-world","strengthening-community-resilience-to-covid19-misinformation","responding-to-the-infodemic","how-were-fighting-the-battle-against-the-covid19-infodemic","toxic-information-is-polluting-our-online-environment",[46,13,47,48,49,50,51,52,53,54,55,56,57,58,59],"nandars-story","foreign-interference-resilience-education","the-coronavirus-quiz","level-up","tackling-community-polarization","project-comet","collab","moving-at-the-speed-of-trust","our-medicine-path","voices-from-the-middle-east","how-can-we-empower-new-voices-online","how-can-we-generate-more-empathy-for-vulnerable-people","political-process-monitoring-in-iran","whats-in-the-bun",[61],{"id":62,"uid":63,"url":63,"type":64,"href":65,"tags":66,"first_publication_date":67,"last_publication_date":68,"slugs":69,"linked_documents":71,"lang":72,"alternate_languages":73,"data":74},"Xo9N7xAAACEALQ1g",null,"master_page","https://orgsite.cdn.prismic.io/api/v2/documents/search?ref=avsfIBEAADMA0TgA&q=%5B%5B%3Ad+%3D+at%28document.id%2C+%22Xo9N7xAAACEALQ1g%22%29+%5D%5D",[],"2020-04-09T16:31:58+0000","2024-10-21T20:13:12+0000",[70],"master-page",[],"en-ca",[],{"site_title":75,"webpage_title_template":76,"support_button_text":77,"support_button":78,"support_button_emoji":88,"support_button_emoji_label":89,"contact_button_text":90,"contact_button_link":91,"contact_button_emoji":95,"contact_button_emoji_label":96,"land_ack_link":97,"land_ack_text":105,"land_ack_button_emoji":106,"land_ack_button_emoji_label":107,"header_navigation":108,"social_media_links":159},"Digital Public Square","%s | Digital Public Square","Support Us",{"id":79,"type":80,"tags":81,"lang":72,"slug":82,"first_publication_date":83,"last_publication_date":84,"link_type":85,"key":86,"isBroken":87},"Xp9e5REAACUAn9Cj","support",[],"support-us","2020-04-21T21:00:25+0000","2020-08-07T20:47:29+0000","Document","5cde7c47-d4d8-455b-bbd7-777b35e455f7",false,"❤️","Heart","Connect With Us",{"link_type":92,"key":93,"url":94},"Web","eb64879a-e08d-48c7-9832-a8a83142edea","https://linktr.ee/digitalpublicsquare","👋","Wave!",{"id":98,"type":99,"tags":100,"lang":72,"slug":101,"first_publication_date":102,"last_publication_date":103,"uid":101,"link_type":85,"key":104,"isBroken":87},"Y_5oQhAAACUAaGiQ","basic_page",[],"land-acknowledgement","2023-02-28T20:47:09+0000","2025-06-04T20:43:54+0000","6740464b-5f17-4c6c-8776-067a30fcedea","Land Acknowledgement","⛰️","Land",[109,119,129,139,149],{"header_link_text":110,"header_link":111},"Home",{"id":112,"type":113,"tags":114,"lang":72,"slug":113,"first_publication_date":115,"last_publication_date":116,"url":117,"link_type":85,"key":118,"isBroken":87},"Xo9ynxAAAB8ALbJn","homepage",[],"2020-04-09T19:08:36+0000","2025-09-16T16:57:32+0000","/","afae4f8b-ae1d-48bb-81ed-3f046b3ba37f",{"header_link_text":120,"header_link":121},"Approach",{"id":122,"type":123,"tags":124,"lang":72,"slug":125,"first_publication_date":126,"last_publication_date":127,"link_type":85,"key":128,"isBroken":87},"Xp9eiREAACQAn88C","approach",[],"approach-page","2020-04-21T20:58:58+0000","2024-10-17T12:38:37+0000","2b629cdc-55f2-417f-b8eb-20efcdbb9f77",{"header_link_text":130,"header_link":131},"Work",{"id":132,"type":133,"tags":134,"lang":72,"slug":135,"first_publication_date":136,"last_publication_date":137,"link_type":85,"key":138,"isBroken":87},"XsQuPxIAACQAMsgy","our_work",[],"our-work","2020-05-19T19:07:31+0000","2020-08-06T20:18:07+0000","3390e854-068a-46e1-a78e-a1417331a4d6",{"header_link_text":140,"header_link":141},"Team",{"id":142,"type":143,"tags":144,"lang":72,"slug":145,"first_publication_date":146,"last_publication_date":147,"link_type":85,"key":148,"isBroken":87},"Xp9eEhEAACQAn8zZ","about",[],"about-page","2020-04-21T20:57:14+0000","2024-06-28T16:27:56+0000","61f43766-c02b-4d44-8dfd-33eaa69947c3",{"header_link_text":150,"header_link":151},"Insights",{"id":152,"type":153,"tags":154,"lang":72,"slug":155,"first_publication_date":156,"last_publication_date":157,"link_type":85,"key":158,"isBroken":87},"X2u-PhAAACMAqErD","blog_list",[],"blog-list","2020-09-23T21:29:41+0000","2025-08-28T13:29:42+0000","71cc4ce3-fc0f-45e3-b478-5a847e7fe3f3",[160,165,171],{"platform":161,"link_to_profile":162,"link_text":161},"Facebook",{"link_type":92,"key":163,"url":164},"5ffb5b43-6e50-4d99-88c2-6b5bd69c89f1","https://www.facebook.com/digitalpublicsquare",{"platform":166,"link_to_profile":167,"link_text":170},"Twitter",{"link_type":92,"key":168,"url":169},"1b9a571f-678d-4c20-b937-abe0aaf49286","https://x.com/dpsorg","X",{"platform":172,"link_to_profile":173,"link_text":172},"LinkedIn",{"link_type":92,"key":174,"url":175},"216c7bc0-549c-4e3d-8834-b9288e461fbc","https://www.linkedin.com/company/digitalpublicsquare/",[61],{"id":178,"uid":4,"url":63,"type":179,"href":180,"tags":181,"first_publication_date":182,"last_publication_date":182,"slugs":183,"linked_documents":185,"lang":72,"alternate_languages":186,"data":187},"avc78RIAACwAOr3o","blog_post","https://orgsite.cdn.prismic.io/api/v2/documents/search?ref=avsfIBEAADMA0TgA&q=%5B%5B%3Ad+%3D+at%28document.id%2C+%22avc78RIAACwAOr3o%22%29+%5D%5D",[],"2026-11-16T15:20:00+0000",[184],"introducing-the-elevate-ai-safety-benchmark-for-preventing-radicalization-to-violence",[],[],{"title":188,"date_posted":194,"author":195,"twitter_handle":63,"author_link":196,"main_image":199,"main_image_description":63,"body":210,"description_snippet":265,"meta_image":266},[189],{"type":190,"text":191,"spans":192,"direction":193},"heading1","Introducing the Elevate AI Safety Benchmark for Preventing Radicalization to Violence",[],"ltr","2026-11-16","Benjamin Windeler and Mackenzie Rice",[197],{"author1":198},{"link_type":85},{"dimensions":200,"alt":203,"copyright":63,"url":204,"id":205,"edit":206},{"width":201,"height":202},1024,680,"AI chatbot landing page that asks \"Ready to go down the rabbit hole?\"","https://images.prismic.io/orgsite/L2WGzzORKiwj2ekk_AIBenchmark-Greeting8.png?auto=format,compress","L2WGzzORKiwj2ekk",{"x":207,"y":207,"zoom":208,"background":209},0,1,"transparent",[211],{"primary":212,"items":262,"id":263,"slice_type":264,"slice_label":63},{"text":213},[214,226,229,232,235,238,241,244,251,254,257],{"type":215,"text":216,"spans":217,"direction":193},"paragraph","Access the Elevate benchmark and technical report here.",[218,221],{"start":207,"end":219,"type":220},55,"em",{"start":207,"end":219,"type":222,"data":223},"hyperlink",{"link_type":92,"url":224,"target":225},"https://www.christchurchcall.org/elevate-benchmark/2026-11/leaderboard","_blank",{"type":215,"text":227,"spans":228,"direction":193},"The violence prevention field is grappling with a range of risks at the intersection of AI safety, terrorism, and violent extremism. Many of these risks involve the exploitation of AI capabilities for attack planning, weapons development, propaganda, and agentic attacks.",[],{"type":215,"text":230,"spans":231,"direction":193},"However, AI safety for preventing terrorism and violent extremism must address more than the exploitation of AI for planning and carrying out attacks – it must also tackle motivation to cause harm. The field needs to broaden from preventing operational support for violence to understanding how system design can shape pathways of radicalization to violence.",[],{"type":215,"text":233,"spans":234,"direction":193},"AI systems are becoming part of how people process information, seek affirmation, and construct meaning. If we develop evaluations that measure only capability-related risks, we will miss where many pathways to violence begin. Mass shootings in Tumbler Ridge and at Florida State University have highlighted the tragedy that can occur when propensity for violence or radicalization signals are missed or met with insufficient responses from AI chatbots.",[],{"type":215,"text":236,"spans":237,"direction":193},"To address this gap, Digital Public Square and the Christchurch Call Foundation developed the Elevate benchmark. The Elevate benchmark is a civil-society led initiative to create a publicly available measure of radicalization to violence risks from AI chatbots. This first iteration tested 11 different models in 124 scenarios, each testing responses in different contexts and trajectories. We worked with a panel of experts working in violence prevention and the broader Christchurch Call community to ensure that the scenarios and evaluation criteria matched aligned with their expertise and standards.",[],{"type":215,"text":239,"spans":240,"direction":193},"We found that while some frontier models performed well across a wide range of scenarios, others did significantly worse with clear room for improvement. Models from Meta, Anthropic, and OpenAI scored highest in evaluations that measured whether models avoided ideological escalation and refused risky requests. Google’s Gemini models and the open weight models that we tested performed much worse.",[],{"type":215,"text":242,"spans":243,"direction":193},"The top performers show clear evidence that AI models are capable of recognizing common risk signals. We believe that all models should be held to a standard where these risks are handled according to best practices.",[],{"type":215,"text":245,"spans":246,"direction":193},"Visit the Elevate AI Safety Benchmark website to see the full leaderboard, analysis, and recommendations, as well as a technical report on the benchmark’s methodology and results.",[247],{"start":248,"end":249,"type":222,"data":250},10,45,{"link_type":92,"url":224,"target":225},{"type":215,"text":252,"spans":253,"direction":193},"At Digital Public Square, it’s our mission to create safer and more resilient digital spaces. We see AI technology as an evolution bringing both new challenges and new opportunities. Careful research into AI harms that recognize human perspectives and impacts will be an essential way that we help society shape this technology into something that strengthens our democratic life.",[],{"type":215,"text":255,"spans":256,"direction":193},"Reach out to hello@digitalpublicsquare.org if you are interested in learning more about this project, or if you want to support our work in helping people navigate democratic risks from digital harms.",[],{"type":215,"text":216,"spans":258,"direction":193},[259,260],{"start":207,"end":219,"type":220},{"start":207,"end":219,"type":222,"data":261},{"link_type":92,"url":224,"target":225},[],"text$606b4cce-35ca-4035-bb64-d66c4b175ef5","text","See the results of our new AI safety benchmark for assessing radicalization to violence risks.",{"dimensions":267,"alt":203,"copyright":63,"url":204,"id":205,"edit":268},{"width":201,"height":202},{"x":207,"y":207,"zoom":208,"background":209},1794842674813]