{"id":42526,"date":"2026-05-26T13:38:49","date_gmt":"2026-05-26T13:38:49","guid":{"rendered":"https:\/\/direc.dk\/secmas-security-in-multi-agent-system-communication\/"},"modified":"2026-06-02T09:52:34","modified_gmt":"2026-06-02T09:52:34","slug":"secmas-security-in-multi-agent-system-communication","status":"publish","type":"post","link":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/","title":{"rendered":"SecMAS \u2013 Security in Multi\u2011Agent System Communication"},"content":{"rendered":"\t\t
\n\t\t\t\t
\n\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

DIREC-projekt<\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

SecMAS \u2013 Security in Multi\u2011Agent System Communication<\/h1>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Projektv\u00e6rdi<\/h2>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tEfterh\u00e5nden som agentiske AI-systemer i stigende grad fungerer som multi-agent-systemer (MAS), opst\u00e5r sikkerhedsrisici ikke l\u00e6ngere kun fra enkelte modeloutputs, men ogs\u00e5 fra m\u00e5den information, autoritet og kontekst bev\u00e6ger sig mellem agenter p\u00e5. Inputs fra ikke-betroede kilder kan sprede sig gennem roller, opsummeringer, delt hukommelse og delegerede v\u00e6rkt\u00f8jskald og potentielt p\u00e5virke handlinger med h\u00f8je privilegier l\u00e6nge efter den oprindelige interaktion.

SecMAS adresserer denne udfordring ved at sikre kommunikationslaget mellem agenter, v\u00e6rkt\u00f8jer og delt kontekst. Ved at g\u00f8re tillidsgr\u00e6nser tydelige og h\u00e5ndh\u00e6velige reducerer projektet risici som k\u00e6dereaktioner af tillidsbrud, uautoriseret brug af v\u00e6rkt\u00f8jer og p\u00e5virkning p\u00e5 tv\u00e6rs af klynger. Dette er s\u00e6rligt vigtigt i h\u00f8jrisikomilj\u00f8er som automatiserede red-team-klynger til penetrationstest, hvor forh\u00f8jede privilegier, f\u00f8lsomme data og strenge krav til isolation skal fungere sammen med effektiv koordinering mellem agenter. SecMAS bidrager til at sikre, at \u00f8get automatisering ikke f\u00f8rer til st\u00f8rre systemisk s\u00e5rbarhed.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

PROJEKTDATA<\/h2>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
Projektnavn<\/strong><\/h5>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

SecMAS \u2013 Security in Multi\u2011Agent System Communication<\/p>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t

\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
Projektperiode<\/strong><\/h5>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t2026-2027\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
Finansiering<\/strong><\/h5>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tDKK 360.000\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Videnskabelig tilgang<\/h2>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tSecMAS fremmer forskning i krydsfeltet mellem protokolsikkerhed, distribuerede systemer og agentbaseret AI ved at definere og h\u00e5ndh\u00e6ve klare og testbare sikkerhedsegenskaber for kommunikation i multi-agent-systemer.

Ved at anvende ColleaiQ\u2019s neuro-symbolske orkestreringsplatform som testmilj\u00f8 udvikler projektet en sikker kommunikationsgateway, der omfatter identitet og autentifikation, autorisation baseret p\u00e5 mindst mulige privilegier, kontrolleret delegering med tidsbegr\u00e6nset eskalering, verifikation af integriteten af beskeder og v\u00e6rkt\u00f8jsoutput samt streng isolering af kontekst.

Sikkerheden h\u00e5ndh\u00e6ves ved en tydelig beslutningsgr\u00e6nse: Inputs fra ikke-betroede kilder kan underst\u00f8tte analyser med lave privilegier, men privilegerede handlinger kr\u00e6ver eksplicitte, policy-godkendte og auditerbare overgange. Forskningen validerer disse mekanismer gennem adversarial testning p\u00e5 faste cybersikkerheds-workflows, hvor b\u00e5de sikkerhedsresultater (s\u00e5som angrebssucces, eskalering og spredningsdybde) og operationelle afvejninger (s\u00e5som latenstid og effektivitet) m\u00e5les. Ved at levere konkrete prototyper, benchmarks og m\u00e5lbar evidens skaber SecMAS et videnskabeligt fundament for en sikrere implementering af agentisk AI i danske og europ\u00e6iske h\u00f8jrisiko-operationelle milj\u00f8er.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Projektdeltagere<\/h2>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n \n \n \n \n \n \n \n \n
\n
\"Nicola<\/div>\n <\/td>\n
\n
Nicola Dragoni – Professor – Technical University of Denmark<\/div>\n <\/td>\n <\/tr>\n \n
\n
\"Christoffer<\/div>\n <\/td>\n
\n
Christoffer Dreist – Co-Founder – ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n
\n
\"Martin<\/div>\n <\/td>\n
\n
Martin Christoffersen – CTO – ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n
\n
\"Mikkel<\/div>\n <\/td>\n
\n
Mikkel Romvig Gr\u00f8ng\u00e5rd – Head of Research – ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n
\n
\"Rasmus<\/div>\n <\/td>\n
\n
Rasmus S\u00f8rensen – ML Engineer – ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n <\/tbody>\n <\/table>\n \t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t

Partnere<\/h2>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\n
\n
\n \"Technical\"ColleaiQ\n <\/div>\n <\/div>\n \t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t","protected":false},"excerpt":{"rendered":"

M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.<\/p>\n","protected":false},"author":3,"featured_media":42533,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[210,218],"tags":[194],"class_list":["post-42526","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-cybersikkerhed","category-projekt","tag-next-gen-cybersecurity-da"],"yoast_head":"\nSecMAS \u2013 Security in Multi\u2011Agent System Communication - DIREC<\/title>\n<meta name=\"description\" content=\"M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/\" \/>\n<meta property=\"og:locale\" content=\"da_DK\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"SecMAS \u2013 Security in Multi\u2011Agent System Communication - DIREC\" \/>\n<meta property=\"og:description\" content=\"M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/\" \/>\n<meta property=\"og:site_name\" content=\"DIREC\" \/>\n<meta property=\"article:published_time\" content=\"2026-05-26T13:38:49+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-06-02T09:52:34+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Agentic-AI.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1920\" \/>\n\t<meta property=\"og:image:height\" content=\"1080\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"Susanne Br\u00f8ndberg\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Skrevet af\" \/>\n\t<meta name=\"twitter:data1\" content=\"Susanne Br\u00f8ndberg\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/\"},\"author\":{\"name\":\"Susanne Br\u00f8ndberg\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#\\\/schema\\\/person\\\/3fcdff2b9bc83fcb7b642fa67a946f38\"},\"headline\":\"SecMAS \u2013 Security in Multi\u2011Agent System Communication\",\"datePublished\":\"2026-05-26T13:38:49+00:00\",\"dateModified\":\"2026-06-02T09:52:34+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/\"},\"wordCount\":32,\"publisher\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/direc.dk\\\/wp-content\\\/uploads\\\/2026\\\/05\\\/Agentic-AI.jpg\",\"keywords\":[\"Next gen cybersecurity\"],\"articleSection\":[\"Cybersikkerhed\",\"Projekt\"],\"inLanguage\":\"da-DK\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/\",\"url\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/\",\"name\":\"SecMAS \u2013 Security in Multi\u2011Agent System Communication - DIREC\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/direc.dk\\\/wp-content\\\/uploads\\\/2026\\\/05\\\/Agentic-AI.jpg\",\"datePublished\":\"2026-05-26T13:38:49+00:00\",\"dateModified\":\"2026-06-02T09:52:34+00:00\",\"description\":\"M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#breadcrumb\"},\"inLanguage\":\"da-DK\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"da-DK\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#primaryimage\",\"url\":\"https:\\\/\\\/direc.dk\\\/wp-content\\\/uploads\\\/2026\\\/05\\\/Agentic-AI.jpg\",\"contentUrl\":\"https:\\\/\\\/direc.dk\\\/wp-content\\\/uploads\\\/2026\\\/05\\\/Agentic-AI.jpg\",\"width\":1920,\"height\":1080},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/secmas-security-in-multi-agent-system-communication\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/direc.dk\\\/da\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"SecMAS \u2013 Security in Multi\u2011Agent System Communication\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#website\",\"url\":\"https:\\\/\\\/direc.dk\\\/da\\\/\",\"name\":\"DIREC\",\"description\":\"Digital Research Centre Denmark\",\"publisher\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/direc.dk\\\/da\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"da-DK\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#organization\",\"name\":\"DIREC\",\"url\":\"https:\\\/\\\/direc.dk\\\/da\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"da-DK\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/direc.dk\\\/wp-content\\\/uploads\\\/2021\\\/02\\\/DIREC_logo_2023_bla.png\",\"contentUrl\":\"https:\\\/\\\/direc.dk\\\/wp-content\\\/uploads\\\/2021\\\/02\\\/DIREC_logo_2023_bla.png\",\"width\":2786,\"height\":786,\"caption\":\"DIREC\"},\"image\":{\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#\\\/schema\\\/logo\\\/image\\\/\"}},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/direc.dk\\\/da\\\/#\\\/schema\\\/person\\\/3fcdff2b9bc83fcb7b642fa67a946f38\",\"name\":\"Susanne Br\u00f8ndberg\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"da-DK\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/e8d2e418b1a0402f1ebb3e45dc93b7b681ec163ec692e71cb65c848ff80d3ab7?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/e8d2e418b1a0402f1ebb3e45dc93b7b681ec163ec692e71cb65c848ff80d3ab7?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/e8d2e418b1a0402f1ebb3e45dc93b7b681ec163ec692e71cb65c848ff80d3ab7?s=96&d=mm&r=g\",\"caption\":\"Susanne Br\u00f8ndberg\"},\"url\":\"https:\\\/\\\/direc.dk\\\/da\\\/author\\\/susanne\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"SecMAS \u2013 Security in Multi\u2011Agent System Communication - DIREC","description":"M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/","og_locale":"da_DK","og_type":"article","og_title":"SecMAS \u2013 Security in Multi\u2011Agent System Communication - DIREC","og_description":"M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.","og_url":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/","og_site_name":"DIREC","article_published_time":"2026-05-26T13:38:49+00:00","article_modified_time":"2026-06-02T09:52:34+00:00","og_image":[{"width":1920,"height":1080,"url":"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Agentic-AI.jpg","type":"image\/jpeg"}],"author":"Susanne Br\u00f8ndberg","twitter_card":"summary_large_image","twitter_misc":{"Skrevet af":"Susanne Br\u00f8ndberg"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#article","isPartOf":{"@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/"},"author":{"name":"Susanne Br\u00f8ndberg","@id":"https:\/\/direc.dk\/da\/#\/schema\/person\/3fcdff2b9bc83fcb7b642fa67a946f38"},"headline":"SecMAS \u2013 Security in Multi\u2011Agent System Communication","datePublished":"2026-05-26T13:38:49+00:00","dateModified":"2026-06-02T09:52:34+00:00","mainEntityOfPage":{"@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/"},"wordCount":32,"publisher":{"@id":"https:\/\/direc.dk\/da\/#organization"},"image":{"@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#primaryimage"},"thumbnailUrl":"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Agentic-AI.jpg","keywords":["Next gen cybersecurity"],"articleSection":["Cybersikkerhed","Projekt"],"inLanguage":"da-DK"},{"@type":"WebPage","@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/","url":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/","name":"SecMAS \u2013 Security in Multi\u2011Agent System Communication - DIREC","isPartOf":{"@id":"https:\/\/direc.dk\/da\/#website"},"primaryImageOfPage":{"@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#primaryimage"},"image":{"@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#primaryimage"},"thumbnailUrl":"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Agentic-AI.jpg","datePublished":"2026-05-26T13:38:49+00:00","dateModified":"2026-06-02T09:52:34+00:00","description":"M\u00e5let med SecMAS er at udvikle og validere sikre kommunikations- og kontrolmekanismer til multi-agent AI-systemer, s\u00e5 uautoriseret adgang, privilegieeskalering og spredning af skadelige inputs mellem agenter kan forhindres. Projektet skal skabe et videnskabeligt fundament for sikker anvendelse af agentisk AI i h\u00f8jrisikomilj\u00f8er som cybersikkerhed og automatiseret penetrationstest.","breadcrumb":{"@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#breadcrumb"},"inLanguage":"da-DK","potentialAction":[{"@type":"ReadAction","target":["https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/"]}]},{"@type":"ImageObject","inLanguage":"da-DK","@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#primaryimage","url":"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Agentic-AI.jpg","contentUrl":"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Agentic-AI.jpg","width":1920,"height":1080},{"@type":"BreadcrumbList","@id":"https:\/\/direc.dk\/da\/secmas-security-in-multi-agent-system-communication\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/direc.dk\/da\/"},{"@type":"ListItem","position":2,"name":"SecMAS \u2013 Security in Multi\u2011Agent System Communication"}]},{"@type":"WebSite","@id":"https:\/\/direc.dk\/da\/#website","url":"https:\/\/direc.dk\/da\/","name":"DIREC","description":"Digital Research Centre Denmark","publisher":{"@id":"https:\/\/direc.dk\/da\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/direc.dk\/da\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"da-DK"},{"@type":"Organization","@id":"https:\/\/direc.dk\/da\/#organization","name":"DIREC","url":"https:\/\/direc.dk\/da\/","logo":{"@type":"ImageObject","inLanguage":"da-DK","@id":"https:\/\/direc.dk\/da\/#\/schema\/logo\/image\/","url":"https:\/\/direc.dk\/wp-content\/uploads\/2021\/02\/DIREC_logo_2023_bla.png","contentUrl":"https:\/\/direc.dk\/wp-content\/uploads\/2021\/02\/DIREC_logo_2023_bla.png","width":2786,"height":786,"caption":"DIREC"},"image":{"@id":"https:\/\/direc.dk\/da\/#\/schema\/logo\/image\/"}},{"@type":"Person","@id":"https:\/\/direc.dk\/da\/#\/schema\/person\/3fcdff2b9bc83fcb7b642fa67a946f38","name":"Susanne Br\u00f8ndberg","image":{"@type":"ImageObject","inLanguage":"da-DK","@id":"https:\/\/secure.gravatar.com\/avatar\/e8d2e418b1a0402f1ebb3e45dc93b7b681ec163ec692e71cb65c848ff80d3ab7?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/e8d2e418b1a0402f1ebb3e45dc93b7b681ec163ec692e71cb65c848ff80d3ab7?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/e8d2e418b1a0402f1ebb3e45dc93b7b681ec163ec692e71cb65c848ff80d3ab7?s=96&d=mm&r=g","caption":"Susanne Br\u00f8ndberg"},"url":"https:\/\/direc.dk\/da\/author\/susanne\/"}]}},"dynamic_page_title":"","dynamic_page_description":"Efterh\u00e5nden som agentiske AI-systemer i stigende grad fungerer som multi-agent-systemer (MAS), opst\u00e5r sikkerhedsrisici ikke l\u00e6ngere kun fra enkelte modeloutputs, men ogs\u00e5 fra m\u00e5den information, autoritet og kontekst bev\u00e6ger sig mellem agenter p\u00e5. Inputs fra ikke-betroede kilder kan sprede sig gennem roller, opsummeringer, delt hukommelse og delegerede v\u00e6rkt\u00f8jskald og potentielt p\u00e5virke handlinger med h\u00f8je privilegier l\u00e6nge efter den oprindelige interaktion.<br><br>SecMAS adresserer denne udfordring ved at sikre kommunikationslaget mellem agenter, v\u00e6rkt\u00f8jer og delt kontekst. Ved at g\u00f8re tillidsgr\u00e6nser tydelige og h\u00e5ndh\u00e6velige reducerer projektet risici som k\u00e6dereaktioner af tillidsbrud, uautoriseret brug af v\u00e6rkt\u00f8jer og p\u00e5virkning p\u00e5 tv\u00e6rs af klynger. Dette er s\u00e6rligt vigtigt i h\u00f8jrisikomilj\u00f8er som automatiserede red-team-klynger til penetrationstest, hvor forh\u00f8jede privilegier, f\u00f8lsomme data og strenge krav til isolation skal fungere sammen med effektiv koordinering mellem agenter. SecMAS bidrager til at sikre, at \u00f8get automatisering ikke f\u00f8rer til st\u00f8rre systemisk s\u00e5rbarhed.","dynamic_page_partners":"\n <div style=\"text-align: left;margin: 20px 0\">\n <div style=\"max-width: 100%\">\n <img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2024\/05\/DTU-logo-lille.png\" alt=\"Technical University of Denmark logo\" style=\"height: 75px;width: auto;min-width: 75px;max-width: 200px;object-fit: contain;margin: 0 25px 25px 0;vertical-align: middle\" \/><img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/ColleaiQ_logo.jpg\" alt=\"ColleaiQ logo\" style=\"height: 75px;width: auto;min-width: 75px;max-width: 200px;object-fit: contain;margin: 0 25px 25px 0;vertical-align: middle\" \/>\n <\/div>\n <\/div>\n ","dynamic_page_scientific_impact":"SecMAS fremmer forskning i krydsfeltet mellem protokolsikkerhed, distribuerede systemer og agentbaseret AI ved at definere og h\u00e5ndh\u00e6ve klare og testbare sikkerhedsegenskaber for kommunikation i multi-agent-systemer.<br><br>Ved at anvende ColleaiQ\u2019s neuro-symbolske orkestreringsplatform som testmilj\u00f8 udvikler projektet en sikker kommunikationsgateway, der omfatter identitet og autentifikation, autorisation baseret p\u00e5 mindst mulige privilegier, kontrolleret delegering med tidsbegr\u00e6nset eskalering, verifikation af integriteten af beskeder og v\u00e6rkt\u00f8jsoutput samt streng isolering af kontekst.<br><br>Sikkerheden h\u00e5ndh\u00e6ves ved en tydelig beslutningsgr\u00e6nse: Inputs fra ikke-betroede kilder kan underst\u00f8tte analyser med lave privilegier, men privilegerede handlinger kr\u00e6ver eksplicitte, policy-godkendte og auditerbare overgange. Forskningen validerer disse mekanismer gennem adversarial testning p\u00e5 faste cybersikkerheds-workflows, hvor b\u00e5de sikkerhedsresultater (s\u00e5som angrebssucces, eskalering og spredningsdybde) og operationelle afvejninger (s\u00e5som latenstid og effektivitet) m\u00e5les. Ved at levere konkrete prototyper, benchmarks og m\u00e5lbar evidens skaber SecMAS et videnskabeligt fundament for en sikrere implementering af agentisk AI i danske og europ\u00e6iske h\u00f8jrisiko-operationelle milj\u00f8er.","dynamic_page_people":"\n <table style=\"width: 100%;border-collapse: collapse;margin: 20px 0;border: none;background: none\">\n <tbody style=\"border: none;background: none\">\n \n <tr>\n <td style=\"padding: 12px;text-align: center;vertical-align: top;border: none;background: none\">\n <div class=\"researcher-photo\"><img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2025\/10\/Nicola-Dragoni.jpg\" alt=\"Nicola Dragoni\" style=\"width: 80px;height: 80px;min-width: 80px;min-height: 80px;border-radius: 50%;object-fit: cover;border: 2px solid #ddd;display: block;margin: 0 auto;max-width: 100%\" \/><\/div>\n <\/td>\n <td style=\"padding: 12px;vertical-align: top;border: none;background: none\">\n <div class=\"participant-name\" style=\"font-weight: normal;font-size: 18px;color: #1C2966\">Nicola Dragoni - Professor - Technical University of Denmark<\/div>\n <\/td>\n <\/tr>\n \n <tr>\n <td style=\"padding: 12px;text-align: center;vertical-align: top;border: none;background: none\">\n <div class=\"researcher-photo\"><img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Christoffer-Dreist.jpg\" alt=\"Christoffer Dreist\" style=\"width: 80px;height: 80px;min-width: 80px;min-height: 80px;border-radius: 50%;object-fit: cover;border: 2px solid #ddd;display: block;margin: 0 auto;max-width: 100%\" \/><\/div>\n <\/td>\n <td style=\"padding: 12px;vertical-align: top;border: none;background: none\">\n <div class=\"participant-name\" style=\"font-weight: normal;font-size: 18px;color: #1C2966\">Christoffer Dreist - Co-Founder - ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n <tr>\n <td style=\"padding: 12px;text-align: center;vertical-align: top;border: none;background: none\">\n <div class=\"researcher-photo\"><img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Martin-Christoffersen.jpg\" alt=\"Martin Christoffersen\" style=\"width: 80px;height: 80px;min-width: 80px;min-height: 80px;border-radius: 50%;object-fit: cover;border: 2px solid #ddd;display: block;margin: 0 auto;max-width: 100%\" \/><\/div>\n <\/td>\n <td style=\"padding: 12px;vertical-align: top;border: none;background: none\">\n <div class=\"participant-name\" style=\"font-weight: normal;font-size: 18px;color: #1C2966\">Martin Christoffersen - CTO - ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n <tr>\n <td style=\"padding: 12px;text-align: center;vertical-align: top;border: none;background: none\">\n <div class=\"researcher-photo\"><img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Mikkel-Romvig.jpg\" alt=\"Mikkel Romvig Gr\u00f8ng\u00e5rd \" style=\"width: 80px;height: 80px;min-width: 80px;min-height: 80px;border-radius: 50%;object-fit: cover;border: 2px solid #ddd;display: block;margin: 0 auto;max-width: 100%\" \/><\/div>\n <\/td>\n <td style=\"padding: 12px;vertical-align: top;border: none;background: none\">\n <div class=\"participant-name\" style=\"font-weight: normal;font-size: 18px;color: #1C2966\">Mikkel Romvig Gr\u00f8ng\u00e5rd - Head of Research - ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n <tr>\n <td style=\"padding: 12px;text-align: center;vertical-align: top;border: none;background: none\">\n <div class=\"researcher-photo\"><img src=\"https:\/\/direc.dk\/wp-content\/uploads\/2026\/05\/Rasmus-Soerensen.jpg\" alt=\"Rasmus S\u00f8rensen\" style=\"width: 80px;height: 80px;min-width: 80px;min-height: 80px;border-radius: 50%;object-fit: cover;border: 2px solid #ddd;display: block;margin: 0 auto;max-width: 100%\" \/><\/div>\n <\/td>\n <td style=\"padding: 12px;vertical-align: top;border: none;background: none\">\n <div class=\"participant-name\" style=\"font-weight: normal;font-size: 18px;color: #1C2966\">Rasmus S\u00f8rensen - ML Engineer - ColleaiQ<\/div>\n <\/td>\n <\/tr>\n \n <\/tbody>\n <\/table>\n ","dynamic_page_project_period":"2026-2027","dynamic_page_funding":"DKK 360.000","dynamic_page_publications":"","dynamic_page_cover_image":"","_links":{"self":[{"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/posts\/42526","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/comments?post=42526"}],"version-history":[{"count":3,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/posts\/42526\/revisions"}],"predecessor-version":[{"id":42571,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/posts\/42526\/revisions\/42571"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/media\/42533"}],"wp:attachment":[{"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/media?parent=42526"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/categories?post=42526"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/direc.dk\/da\/wp-json\/wp\/v2\/tags?post=42526"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}