{"id":5287,"date":"2026-02-24T04:14:15","date_gmt":"2026-02-24T04:14:15","guid":{"rendered":"https:\/\/research.undetectable.ai\/?p=5287"},"modified":"2026-02-24T04:14:50","modified_gmt":"2026-02-24T04:14:50","slug":"i-tested-5-ai-image-detectors-these-are-the-best-in-2026","status":"publish","type":"post","link":"https:\/\/research.undetectable.ai\/nb\/jeg-testet-5-ai-bildedetektorer-disse-er-de-beste-i-2026\/","title":{"rendered":"Jeg testet 5 AI-bildedetektorer. Disse er de beste i 2026:"},"content":{"rendered":"

Du har sannsynligvis sett flere AI-genererte bilder p\u00e5 nettet enn du er klar over. <\/strong><\/p>\n\n\n\n

Noen ganger er det \u00e5penbart at et bilde er generert av kunstig intelligens, men det blir stadig vanskeligere \u00e5 se det etter hvert som de generative bilde- og videoverkt\u00f8yene blir bedre. Nye verkt\u00f8y som Googles Nano Banana Pro<\/a>og oppdateringer til OpenAIs ChatGPT-bildemodell<\/a> lar brukerne raskt generere syntetiske bilder som speiler virkelige bilder. Tidligere unders\u00f8kelser har vist at 85% av amerikanerne sier deepfakes svekker tilliten p\u00e5 nettet<\/a>.<\/p>\n\n\n\n

Flere verkt\u00f8y hevder \u00e5 kunne oppdage AI-genererte bilder, men fungerer de? <\/strong><\/h2>\n\n\n\n

Jeg kj\u00f8rte 50 deteksjonssjekker p\u00e5 tvers av fem av de mest popul\u00e6re AI-bildedetektorene og dokumenterte resultatene. Ikke bare vil jeg presentere alle dataene i denne artikkelen og forklare dem, jeg vil ogs\u00e5 lenke til dokumentasjonen til slutt.<\/p>\n\n\n\n

De fem detektorene jeg brukte til disse testene, var TruthScan<\/a>, AI eller ikke<\/a>, Siktmotor<\/a>, WasItAI<\/a>, og Winston AI<\/a>. Jeg testet disse verkt\u00f8yene ved \u00e5 la hvert av dem analysere to ChatGPT-genererte bilder, seks Nano Banana-genererte bilder og to bilder generert av Midjourney. Jeg brukte flere promptstiler og -teknikker (som jeg vil g\u00e5 n\u00e6rmere inn p\u00e5 senere n\u00e5r jeg analyserer hvert enkelt bilde).<\/p>\n\n\n\n

Alle de fem detektorene fikk i oppgave \u00e5 oppdage AI-skapte bilder i kategorier som svindel, desinformasjon, generell fotografering og deepfakes. Ikke overraskende var det ikke alle detektorene som presterte godt. TruthScan var den eneste detektoren som konsekvent klassifiserte alt innholdet jeg sendte inn, til 97% eller h\u00f8yere.<\/strong><\/p>\n\n\n\n\n\n\n\n\nTestresultater for AI Image Detector<\/title>\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .ai-detector-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .ai-detector-wrap h2 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.75rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .ai-detector-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .ai-detector-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.85rem;\n text-transform: uppercase;\n letter-spacing: 0.06em;\n padding: 16px 20px;\n text-align: center;\n border: none;\n }\n\n .ai-detector-table thead th:first-child {\n text-align: left;\n padding-left: 24px;\n }\n\n .ai-detector-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .ai-detector-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .ai-detector-table tbody td {\n padding: 14px 20px;\n text-align: center;\n font-size: 0.95rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .ai-detector-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .ai-detector-table tbody td:first-child {\n text-align: left;\n padding-left: 24px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n \/* Colored cells *\/\n .cell-green {\n background: #22c55e;\n color: #fff !important;\n font-weight: 700 !important;\n border-radius: 6px;\n padding: 6px 12px;\n display: inline-block;\n min-width: 32px;\n text-align: center;\n font-size: 0.9rem;\n }\n\n .cell-red {\n background: #ef4444;\n color: #fff !important;\n font-weight: 700 !important;\n border-radius: 6px;\n padding: 6px 12px;\n display: inline-block;\n min-width: 32px;\n text-align: center;\n font-size: 0.9rem;\n }\n\n \/* Accuracy badges *\/\n .badge-100 { background: #16a34a; color: #fff; }\n .badge-80 { background: #a3e635; color: #1a1a2e; }\n .badge-70 { background: #facc15; color: #1a1a2e; }\n .badge-60 { background: #fb923c; color: #fff; }\n .badge-30 { background: #ef4444; color: #fff; }\n\n .accuracy-badge {\n font-weight: 800;\n font-family: 'Outfit', sans-serif;\n font-size: 0.95rem;\n border-radius: 8px;\n padding: 6px 16px;\n display: inline-block;\n min-width: 60px;\n text-align: center;\n letter-spacing: 0.02em;\n }\n\n \/* Row rank indicator *\/\n .rank-icon {\n margin-right: 6px;\n font-size: 1rem;\n }\n\n \/* Asterisk note *\/\n .table-note {\n margin-top: 0.75rem;\n font-size: 0.8rem;\n color: #888;\n padding-left: 4px;\n }\n\n \/* Responsive *\/\n @media (max-width: 600px) {\n .ai-detector-table thead th,\n .ai-detector-table tbody td {\n padding: 10px 10px;\n font-size: 0.8rem;\n }\n .ai-detector-wrap h2 {\n font-size: 1.3rem;\n }\n }\n<\/style>\n<\/head>\n<body style=\"background:#f8f8fa; padding: 2rem;\">\n\n<div class=\"ai-detector-wrap\">\n <h2>\ud83d\udd0d Testresultater for AI Image Detector<\/h2>\n\n <table class=\"ai-detector-table\">\n <thead>\n <tr>\n <th>Deteksjonsverkt\u00f8y<\/th>\n <th>Totalt antall tester<\/th>\n <th>Korrekt detektert<\/th>\n <th>Bommet (mislykkes)<\/th>\n <th>N\u00f8yaktighet<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td><span class=\"rank-icon\">\ud83e\udd47<\/span>TruthScan<\/td>\n <td>10<\/td>\n <td><span class=\"cell-green\">10<\/span><\/td>\n <td><span class=\"cell-green\">0<\/span><\/td>\n <td><span class=\"accuracy-badge badge-100\">100%<\/span><\/td>\n <\/tr>\n <tr>\n <td><span class=\"rank-icon\">\ud83e\udd48<\/span>AI eller ikke<\/td>\n <td>10<\/td>\n <td><span class=\"cell-green\">8<\/span><\/td>\n <td><span class=\"cell-red\">2*<\/span><\/td>\n <td><span class=\"accuracy-badge badge-80\">80%<\/span><\/td>\n <\/tr>\n <tr>\n <td><span class=\"rank-icon\">\ud83e\udd49<\/span>Siktmotor<\/td>\n <td>10<\/td>\n <td><span class=\"cell-green\">7<\/span><\/td>\n <td><span class=\"cell-red\">3<\/span><\/td>\n <td><span class=\"accuracy-badge badge-70\">70%<\/span><\/td>\n <\/tr>\n <tr>\n <td><span class=\"rank-icon\">   <\/span>WasItAI<\/td>\n <td>10<\/td>\n <td><span class=\"cell-green\">6<\/span><\/td>\n <td><span class=\"cell-red\">4<\/span><\/td>\n <td><span class=\"accuracy-badge badge-60\">60%<\/span><\/td>\n <\/tr>\n <tr>\n <td><span class=\"rank-icon\">   <\/span>Winston AI<\/td>\n <td>10<\/td>\n <td><span class=\"cell-green\">3<\/span><\/td>\n <td><span class=\"cell-red\">7<\/span><\/td>\n <td><span class=\"accuracy-badge badge-30\">30%<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n\n <p class=\"table-note\">* AI or Not merket \u00e9n AI-generasjon som 78%, \u00e9n som 85% og \u00e9n som 89%. Siden testen krever minst 90%, teller vi de to f\u00f8rste som ikke best\u00e5tt, men behandler 89% som en nesten-ikke-best\u00e5tt, derav stjernen.<\/p>\n<\/div>\n\n<\/body>\n<\/html>\n\n\n\n<p style=\"font-size:17px\">AI or Not presterte ogs\u00e5 ganske bra. Selv om AI or Not klassifiserte flere AI-genererte gjenstander under 90%-sikkerhet, var det faktisk den nest mest n\u00f8yaktige og konsekvente detektoren under testene mine, etter TruthScan. Selv der AI or Not feilet, var ikke disse feilene katastrofale, men det er fortsatt rom for forbedringer. <\/p>\n\n\n\n<p style=\"font-size:17px\">Resten av AI-bildedetektorene jeg pr\u00f8vde, presterte langt d\u00e5rligere. Sight Engine feilklassifiserte for eksempel tre svindelrelaterte AI-bilder som autentiske.<\/p>\n\n\n\n<p style=\"font-size:17px\">N\u00e5 skal jeg vise deg hvert bilde (av de 10 jeg genererte), forklare hvordan jeg skapte det, og vise hvordan hver modell scoret det.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#1. \"Man On a Ledge\" (generert av ChatGPT)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"683\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-683x1024.jpg\" alt=\"\" class=\"wp-image-5289\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-683x1024.jpg 683w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-200x300.jpg 200w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-768x1152.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge-8x12.jpg 8w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT-Generation-1-Man-on-ledge.jpg 1024w\" sizes=\"auto, (max-width: 683px) 100vw, 683px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 12%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>ChatGPT <span style=\"font-size:0.75rem; color:#888; display:block;\">Full generasjon<\/span><\/td>\n <td>Generelt<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">77.99% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">0,98% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p style=\"font-size:17px\">For meg (og jeg ser mye AI-generert innhold hver dag) ser bildet ganske overbevisende ut ved f\u00f8rste \u00f8yekast. Hvis du klikket p\u00e5 en persons profil p\u00e5 sosiale medier, scrollet gjennom feeden deres og s\u00e5 dette bildet, ville det umiddelbart ha skilt seg ut som en AI-generert forfalskning? For \u00e5 v\u00e6re rettferdig pr\u00f8vde jeg \u00e5 v\u00e6re kreativ med ledeteksten. Her er den jeg brukte til \u00e5 generere dette bildet:<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"has-text-align-center has-vivid-cyan-blue-color has-text-color has-link-color has-medium-font-size wp-elements-99b76fc78211de13a760aff177dd504c\">“<em><strong>Generer et bilde i snapshot-kamerastil av en mann som st\u00e5r p\u00e5 toppen av et tak og ser ned p\u00e5 gaten, nattetid, han er opplyst av kamerablitsen, det er lysspor fra bilene nedenfor, en kvinne st\u00e5r bak ham med h\u00e5nden p\u00e5 leppene og smiler, estetikken i bildet er som et bilde tatt p\u00e5 et snapshot-kamera i 2009 av skolevenner som tuller rundt.<\/strong><\/em>“<\/p>\n<\/blockquote>\n<\/blockquote>\n<\/blockquote>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p style=\"font-size:17px\">Uansett hvor kreativ jeg var eller hvor mange detaljer jeg inkluderte i ledeteksten, klassifiserte TruthScan og Sight Engine likevel utdataene som AI-genererte. AI or Not-detektoren var n\u00e6rme, men bommet likevel p\u00e5 m\u00e5let. <\/p>\n\n\n\n<p style=\"font-size:17px\">Winston og WasItAI tok fullstendig feil da de klassifiserte ChatGPT-bildet som ekte.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#2. \"Den falske kvitteringen\" (generert av ChatGPT)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"683\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-683x1024.jpg\" alt=\"Et bilde som viser det som ser ut til \u00e5 v\u00e6re en flekkete kvittering\" class=\"wp-image-5291\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-683x1024.jpg 683w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-200x300.jpg 200w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-768x1152.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest-8x12.jpg 8w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/CGPT2Bestbuyreceipttest.jpg 1024w\" sizes=\"auto, (max-width: 683px) 100vw, 683px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 12%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>ChatGPT <span style=\"font-size:0.75rem; color:#888; display:block;\">Full generasjon<\/span><\/td>\n <td>Bedrageri<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">19.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">94.48% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">0,04% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Jeg ville generere en falsk kvittering med ChatGPT for \u00e5 se hvor realistisk den ville se ut. Du vil kanskje legge merke til at bildet mangler en legitim adresse, men teksturen ser fortsatt litt realistisk ut. Jeg vil si at dette bildet er mindre overbevisende enn det f\u00f8rste jeg fikk ChatGPT til \u00e5 generere, men de fleste detektorene gikk likevel i bakken da de analyserte det.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li style=\"font-size:17px\">Av alle kontrollene jeg kj\u00f8rte p\u00e5 dette bildet, <strong>TruthScan<\/strong> var den mest n\u00f8yaktige. <\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>Siktmotor<\/strong> virkelig mislyktes her, og viste d\u00e5rlig d\u00f8mmekraft p\u00e5 et dokumentrelatert AI-bilde.<\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>AI eller ikke<\/strong> gjorde det merkbart bedre i denne andre analysen.<\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>Winston AI <\/strong>presterte d\u00e5rligst, og WasItAI mislyktes ogs\u00e5 fullstendig.<\/li>\n<\/ul>\n\n\n\n<p style=\"font-size:17px\">Med tanke p\u00e5 at dette bildet er assosiert med svindelkategorien, er det bekymringsfullt at de fleste detektorene feilklassifiserte det. <\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Prompten jeg brukte til \u00e5 generere bildet<\/h3>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-b58fc16e4a944c203389dba029f958d9\"><strong>“<em>Generer et bilde av en kvittering fra Best Buy der det st\u00e5r at det totale bel\u00f8pet som er brukt er \u00e9n milliard dollar, og at det er en kaffeflekk p\u00e5 den<\/em>“<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p style=\"font-size:17px\">Jeg skrev tydeligvis en rask og enkel melding. Likevel er kvaliteten p\u00e5 ChatGPT-utdataene visuelt overbevisende for det blotte \u00f8ye. Ikke s\u00e5 overbevisende for to av de fem detektorene jeg testet, ser det ut til.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#3: \"Hazard Package\" (generert med Nano Banana)<\/h2>\n\n\n\n<p><strong><em>I stedet for \u00e5 la Nano Banana generere et originalbilde, ga jeg den et bilde \u00e5 endre. <\/em><\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">Tanken var \u00e5 demonstrere hvordan noen enkelt kan bruke kunstig intelligens til \u00e5 skape falske bevis og hevde at de har mottatt en skadet\/farlig pakke. F\u00f8rst tok jeg et bilde med en iPhone 15 av en tom Amazon-pakke jeg fant.<\/p>\n\n\n\n<p><strong>Her er bildet jeg tok:<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"759\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-759x1024.jpg\" alt=\"Et bilde av en Amazon-pakke\" class=\"wp-image-5293\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-759x1024.jpg 759w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-222x300.jpg 222w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-768x1036.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Amazon-package-9x12.jpg 9w\" sizes=\"auto, (max-width: 759px) 100vw, 759px\" \/><\/figure>\n\n\n\n<p style=\"font-size:17px\">Deretter lastet jeg opp bildet jeg tok til Nano Banana og ga den f\u00f8lgende beskjed:<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-41ee9a66efaa63aab6aedf0d2e80cce1\"><strong>“<em>Jeg vil at du skal redigere dette bildet, ikke endre noe bortsett fra der etiketten er p\u00e5, legg til skade p\u00e5 den og legg til svarte slamflekker p\u00e5 esken som ser ekte ut som fettflekker<\/em>“<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">Nano Bana-utgang:<\/h3>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"747\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-747x1024.jpg\" alt=\"Et bilde som viser en pakke\" class=\"wp-image-5292\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-747x1024.jpg 747w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-219x300.jpg 219w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-768x1052.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged-9x12.jpg 9w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Amazon-Package-Damaged.jpg 864w\" sizes=\"auto, (max-width: 747px) 100vw, 747px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 12%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banana <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n <td>Bedrageri<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">98.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.10% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">31.98% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Det oljelignende slammet som Nano Banana genererte i utdataene, hadde en overbevisende skitten glans, komplett med en fuktig bl\u00f8tleggingseffekt - ikke tegneserieaktig i det hele tatt. Heldigvis stemplet de fleste detektorene jeg kj\u00f8rte det gjennom bildet som en forfalskning. Den eneste avvikeren var Winston, som igjen gjorde det d\u00e5rlig og feilklassifiserte bildet som autentisk. AI or Not, TruthScan og WasItAI var de mest n\u00f8yaktige detektorene n\u00e5r det gjaldt \u00e5 oppdage dette bildet.<\/p>\n\n\n\n<p style=\"font-size:17px\">Det er urovekkende hvor enkelt det virker \u00e5 laste opp et bilde til en chatbot og f\u00e5 det fullstendig endret. Jeg kan se for meg at svindlere og bedragere bruker generative bildeverkt\u00f8y for \u00e5 pr\u00f8ve \u00e5 fremsette falske returkrav eller drive med markedssvindel.<\/p>\n\n\n\n<p style=\"font-size:17px\">Hvis e-handelsplattformer bare krever et bilde som bevis for \u00e5 starte et krav om en skadet pakke, uten noen p\u00e5litelig m\u00e5te \u00e5 oppdage AI-manipulering p\u00e5, er de utsatt for en viktig angrepsvektor. For eksempel er Amazons offisielle <a href=\"https:\/\/sellercentral.amazon.com\/help\/hub\/reference\/external\/GYW2EV9FEMJ3JJGR?locale=en-US\" target=\"_blank\" rel=\"noopener\">retningslinjer for retur av varer<\/a> bokstavelig talt sier<strong>, “<em>Farlig gods, inkludert brennbare v\u00e6sker eller gasser, kan ikke returneres<\/em>.”<\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">S\u00e5 hvis du faktisk mottok en pakke dekket av olje eller fettete svart slam, trenger du ikke \u00e5 returnere den, men du kan fortsatt f\u00e5 refusjon. Hvis det av en eller annen grunn faktisk skjedde med deg, ville Amazon sannsynligvis be om bildebevis. Skj\u00f8nner du hvilket problem AI-genererte bilder kan utgj\u00f8re n\u00e5r de ikke kan skilles fra ekte bilder?<\/p>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-left\">#4. \"Kakerlakkm\u00e5ltid\" (generert med Nano Banana)<\/h2>\n\n\n\n<p style=\"font-size:17px\">Dette neste AI-bildet kan f\u00e5 deg til \u00e5 miste appetitten. Det viser seg at det finnes dokumenterte tilfeller av mennesker som bruker kunstig intelligens til \u00e5 motta <a href=\"https:\/\/dailydot.com\/ai-generated-undercooked-food-doordash\" data-type=\"link\" data-id=\"https:\/\/dailydot.com\/ai-generated-undercooked-food-doordash\" target=\"_blank\" rel=\"noopener\">falske refusjoner<\/a> p\u00e5 matleveringsapper. I denne fjerde testen fulgte jeg de samme trinnene som i test tre. Denne gangen tok jeg et bilde av den tomme matboksen min etter lunsj.<\/p>\n\n\n\n<p><strong>Her er det virkelige bildet:<\/strong><\/p>\n\n\n\n<figure class=\"wp-block-image size-large is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"768\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-768x1024.jpg\" alt=\"Et bilde av en tom matkasse.\" class=\"wp-image-5296\" style=\"width:462px;height:auto\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-768x1024.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-225x300.jpg 225w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-scaled.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/IMG_6145-9x12.jpg 9w\" sizes=\"auto, (max-width: 768px) 100vw, 768px\" \/><\/figure>\n\n\n\n<p style=\"font-size:17px\">Deretter la jeg det ekte bildet jeg tok inn i Nano Banana og ga den f\u00f8lgende beskjed:<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-e9f00ca4e893e7742bce463148a639f5\"><strong><em>\"Rediger dette bildet. Ikke endre noe, bortsett fra \u00e5 legge til nudler der folien er, og legge til en haug med sm\u00e5 kakerlakker i maten. F\u00e5 det hele til \u00e5 se realistisk ut og ikke tegneserieaktig.\" <\/em><\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">Nano Banana Output:<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"747\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal.jpg\" alt=\"Et AI-behandlet bilde av takeaway-esken med nudler og insekter inni esken\" class=\"wp-image-5295\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal.jpg 747w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal-219x300.jpg 219w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Cockroach-meal-9x12.jpg 9w\" sizes=\"auto, (max-width: 747px) 100vw, 747px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 12%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banana <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n <td>Bedrageri<\/td>\n <td><span class=\"score-pill score-pass\">97.15% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">18.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">85.87% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">72.29% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Nok en gang oppdaget TruthScan med sikkerhet AI-genererte medier. Sight Engine og WasitAI mislyktes fullstendig. AI or Not var den nest mest n\u00f8yaktige detektoren (med 85% AI-sikkerhet) ved skanning av dette AI-bildet, og Winston presterte bedre enn den gjorde i den tredje testen (med 72% AI-sikkerhet).<\/p>\n\n\n\n<p style=\"font-size:17px\">I dette bildet ble alle nudlene og kakerlakkene generert utelukkende av kunstig intelligens, men to av detektorene s\u00e5 ikke ut til \u00e5 tro det, og bare \u00e9n (TruthScan) var over 90% sikker p\u00e5 at det var kunstig intelligens som hadde tuklet.<\/p>\n\n\n\n<p style=\"font-size:17px\">Et viktig poeng her var hvor enkelt det var \u00e5 lage dette falske bildet. Fra jeg lastet opp bildet jeg tok, til jeg ba Nano Banana om \u00e5 gj\u00f8re det og mottok resultatet, tok det bare rundt to minutter. Et mareritt for matleveringsapper; en dr\u00f8m for sultne l\u00f8gnere. En \"svindelfaktor\".<\/p>\n\n\n\n<p class=\"has-vivid-red-color has-text-color has-link-color wp-elements-4c89b6ee3afcbdd6b6617c150e5b2e03\"><strong>OFFENTLIGHET: Noen av de AI-genererte bildene som inng\u00e5r i de f\u00f8lgende testene, er ment \u00e5 demonstrere hvordan AI-bilder kan brukes til desinformasjon. De inneholder kontroversielle bilder og sensitive emner.<\/strong>, <strong>og er IKKE ment \u00e5 representere noen bestemt politisk<\/strong> <strong>synspunkter eller ideologier.<\/strong><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>#5<\/strong>. \"Autovandalisme\" (generert med Nano Banana)<\/h2>\n\n\n\n<p style=\"font-size:17px\">Jeg fant et ekte bilde av en bil til salgs p\u00e5 nettet. Ganske banalt. Men hva om noen \u00f8nsket \u00e5 bruke dette bildet til \u00e5 spre desinformasjon, eller sende inn et falskt bilforsikringskrav? Dessverre gj\u00f8r generativ AI det raskt og enkelt.<\/p>\n\n\n\n<p style=\"font-size:17px\"><strong>I denne femte testen tok jeg et ekte bilde av en bil, og ga Nano Banana f\u00f8lgende beskjed:<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-d06074fe9533ce61c08bcf9750cd398d\">“<strong><em>Ikke endre noe p\u00e5 bilen, bortsett fra at jeg vil at du skal spraymale den med \"Free Palestine\", og at frontruten er sprukket. Frontlyktene er knust og speilet er \u00f8delagt og henger ned, og de to vinduene p\u00e5 siden av bilen er<\/em><\/strong> <strong>knust<\/strong>.”<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Her er det virkelige bildet:<\/strong><\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"940\" height=\"627\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2.webp\" alt=\"Et bilde av en bl\u00e5 bil\" class=\"wp-image-5298\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2.webp 940w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2-300x200.webp 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2-768x512.webp 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/2000_honda_civic-si_2000_honda_civic-si_bbeed9f1-7b50-4314-a504-40be9411b8ab-mXhf1y-2-18x12.webp 18w\" sizes=\"auto, (max-width: 940px) 100vw, 940px\" \/><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\">Nano Banana AI-bilde:<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"683\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage.jpg\" alt=\"Et AI-behandlet bilde av den bl\u00e5 bilen, som viser en sprukket frontrute og lakkskader p\u00e5 siden av bilen\" class=\"wp-image-5299\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage-300x200.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage-768x512.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Protest-Car-Damage-18x12.jpg 18w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 14%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banana <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n <td style=\"font-size:0.82rem;\">Bedrageri \/ desinformasjon<\/td>\n <td><span class=\"score-pill score-pass\">97.48% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">18.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">89.52% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">0,02% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Detektorresultatene for denne utgangen var blandede. For denne testen hadde WasItAI (overraskende nok) den h\u00f8yeste AI-deteksjonsscore (med 99%). TruthScan hadde den nest h\u00f8yeste poengsummen (97%). Sight Engine klarte ikke \u00e5 oppdage noen syntetiske elementer i bildet, og Winston gjorde det d\u00e5rligst av alle detektorene (med mindre enn 1% sjanse for involvering). AI or Not gjorde det bra, men n\u00f8yaktigheten var fortsatt under 90%.<br><br>La oss snakke om bildet. De to mest bekymringsfulle aspektene ved AI-skapt eller endret innhold som dette, er at begge involverer bedrag. For det f\u00f8rste er det raskt, enkelt og ser ekte ut \u00e5 bruke kunstig intelligens til \u00e5 endre hverdagslige bilder for \u00e5 skape politisk ladet innhold. Enten de som lager denne typen innhold bare er ute etter \u00e5 skape engasjement eller er politiske provokat\u00f8rer, er problemet at det er falsk informasjon som presenteres som autentisk.<\/p>\n\n\n\n<p style=\"font-size:17px\">Den andre m\u00e5ten slike bildegenereringsverkt\u00f8y kan misbrukes p\u00e5, er til svindel. Hvis en svindler \u00f8nsker \u00e5 sende inn et falskt bilforsikringskrav, m\u00e5 de produsere falske bevis. I stedet for \u00e5 bruke timevis i Photoshop p\u00e5 \u00e5 redigere bilder og manipulere bevis, kan de bruke bildegeneratorer til raskt \u00e5 produsere falske bevis.<\/p>\n\n\n\n<p style=\"font-size:17px\">I AI-bildeeksemplet ovenfor kan du se hvordan en svindler kan bruke en lignende prosess for \u00e5 sende inn et falskt krav i henhold til Progressives <a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-vandalism\/\" data-type=\"link\" data-id=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-vandalism\/\" target=\"_blank\" rel=\"noopener\">Omfattende dekningspolicy<\/a>, som spesifikt dekker:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Sk\u00e5ret eller <a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-tire-damage\/\" target=\"_blank\" rel=\"noopener\">\u00f8delagte dekk<\/a><\/strong><\/li>\n\n\n\n<li><strong>\u00d8delagte vinduer, frontlykter eller baklykter<\/strong><\/li>\n\n\n\n<li><strong>Skader fra spraymaling<\/strong><\/li>\n\n\n\n<li><strong><a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-scratches\/\" target=\"_blank\" rel=\"noopener\">Bulker eller riper<\/a> fra at noen stjeler biln\u00f8klene dine<\/strong><\/li>\n\n\n\n<li><strong>Fyller sukker eller andre stoffer i bensintanken<\/strong><\/li>\n<\/ul>\n\n\n\n<p>\u00c5 fremlegge falske bevis er selvsagt ikke bare villedende, det er ogs\u00e5 direkte ulovlig. I en ideell verden ville ingen bryte loven, men i virkeligheten skjer det kriminalitet hver dag. Og uten p\u00e5litelig deteksjon risikerer bilforsikringsselskapene \u00e5 bli svindlet for millioner etter hvert som AI-aktiverte falske forsikringskrav \u00f8ker.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#6. \"Soda Musk and Don\" (generert med Nano Banana)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"559\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-1024x559.jpeg\" alt=\"Et AI-generert bilde av Elon Musk og Donald Trump\" class=\"wp-image-5300\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-1024x559.jpeg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-300x164.jpeg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-768x419.jpeg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-1536x838.jpeg 1536w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-2048x1117.jpeg 1877w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-18x10.jpeg 18w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3-Elon-Spilling-Soda-scaled.jpeg 1878w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 14%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banana <span style=\"font-size:0.75rem; color:#888; display:block;\">Full generasjon<\/span><\/td>\n <td style=\"font-size:0.82rem;\">Disinfo<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">0,24% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">86.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Ogs\u00e5 i denne testen brukte jeg Nano Banana, og m\u00e5let var \u00e5 demonstrere falske, politisk ladede bilder med en litt mer tullete undertone. TruthScan, Sight Engine og AIorNot flagget alle det genererte bildet med en AI-deteksjonsvurdering p\u00e5 99%. WasitAI var ikke like selvsikker - de oppdaget det AI-genererte bildet med bare 86% sikkerhet. Og Winston mislyktes fullstendig, og ga en AI-poengsum p\u00e5 0,2%. For \u00e5 generere bildet ga jeg Nano Banana denne beskjeden:<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p class=\"has-vivid-cyan-blue-color has-text-color has-link-color wp-elements-4f7e002ecaefec24bbebdca3101fff42\"><strong>“<em>Generer et bilde av Elon Musk som s\u00f8ler en stor kopp brus som en Big Gulp p\u00e5 skjorten sin og flipper ut mens Donald Trump sitter ved siden av ham og ler. De er p\u00e5 et jetfly. Det er scenen. Og bildet skal se realistisk ut, som om det er tatt med et snapshot-kamera, og inkludere artefakter som at kamerablitsen reflekteres i flyvinduet og utenfor flyet. Det er natt. Det skal se hyperrealistisk ut.<\/em>“<\/strong><\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p style=\"font-size:17px\">N\u00e5r et av disse programmene ikke klarer \u00e5 oppdage AI, stempler de det som autentisk. Selv om dette bildeeksemplet fra testen er dumt, er det ikke noe morsomt ved \u00e5 tro at en deepfake er ekte.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#7. \"Still Alive\" (generert av Nano Banana)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"765\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo.jpg\" alt=\"Et AI-generert bilde av en mann som holder et skilt med teksten "hjelp"\" class=\"wp-image-5301\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo.jpg 765w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo-224x300.jpg 224w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/GV3Christian_Ransom_Photo-9x12.jpg 9w\" sizes=\"auto, (max-width: 765px) 100vw, 765px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 14%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banana <span style=\"font-size:0.75rem; color:#888; display:block;\">Full generasjon<\/span><\/td>\n <td style=\"font-size:0.82rem;\">Deepfake<\/td>\n <td><span class=\"score-pill score-pass\">97.49% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">95.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.27% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">0,15% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">1.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Synes du dette bildet er urovekkende? For \u00e5 generere denne deepfake brukte jeg Nano Banana og ga den et bilde av meg selv. Deretter ga jeg den en lang ledetekst som skulle sette meg i en situasjon som antydet at jeg hadde blitt bortf\u00f8rt. Kanskje det er fordi dette bildet viser <em>meg<\/em> i en kompromitterende posisjon. Tenk om noen lagde noe s\u00e5nt, sendte det til moren min og truet med \u00e5 skade meg hvis hun ikke sendte dem penger? Skremmende greier. <\/p>\n\n\n\n<p class=\"has-small-font-size\"><em>(Den fullstendige ledeteksten jeg brukte til dette, finner du i det fullstendige testdatasettet p\u00e5 slutten av denne artikkelen).<\/em><\/p>\n\n\n\n<p style=\"font-size:17px\"><br>Heldigvis for dette bildet, til tross for at det har et veldig realistisk utseende, <strong>tre av fem detektorer identifiserte den som AI-generert.<\/strong> <\/p>\n\n\n\n<p style=\"font-size:17px\">AI or Not fikk en sterk AI-score p\u00e5 99%, TruthScan 97% og Sight Engine 95%. Dessverre klassifiserte Winston og WasItAI bildet som ekte.<\/p>\n\n\n\n<p style=\"font-size:17px\">Jeg m\u00e5 presisere at disse testene ikke ble gjort p\u00e5 min personlige Gemini-konto. Nano Banana ba meg aldri om \u00e5 bevise at jeg var personen p\u00e5 bildet. Hvem som helst kunne ha lastet ned et bilde av meg fra internett, lagt det inn i Nano Banana og f\u00e5tt det til \u00e5 skape denne typen bilder. <\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#8. \"Abuse of a Ben\" (generert med Nano Banana)<\/h2>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"200\" height=\"200\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1.jpg\" alt=\"\" class=\"wp-image-5307\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1.jpg 200w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1-150x150.jpg 150w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/c945d528-730f-48c1-bdca-30e5799fc1aa_thumb-1-12x12.jpg 12w\" sizes=\"auto, (max-width: 200px) 100vw, 200px\" \/><\/figure>\n<\/div>\n\n\n<p style=\"font-size:17px\">M\u00f8t Ben. Han tilbringer mesteparten av tiden sin med \u00e5 lede ulike avdelinger i selskapene han jobber i. Bildet over er Bens (ekte) stilige corporate headshot. Hvis du ikke ser det, er Ben en profesjonell fyr. Uansett hvilke politiske meninger han har, holder han dem for seg selv, og samtalene p\u00e5 jobb handler bare om jobb. Men n\u00e5r som helst kan Bens bilde bli brukt til \u00e5 spre politisk propaganda eller budskap uten hans samtykke.<\/p>\n\n\n\n<p style=\"font-size:17px\">Den siste tiden har det v\u00e6rt mye rapportering om hvordan AI-genererte videoer med ICE-agenter og protester har spredd seg p\u00e5 nettet. Basert p\u00e5 alle testene s\u00e5 langt konkluderte jeg med at det var veldig enkelt \u00e5 lage falsk politisk propaganda med kunstig intelligens. Jeg bestemte meg for \u00e5 teste igjen, denne gangen med Ben. <\/p>\n\n\n\n<p style=\"font-size:17px\">Med Bens tillatelse brukte jeg hodeskuddet hans til \u00e5 lage en deepfake. <strong>Her er det AI-genererte bildet som Nano Banana har laget:<\/strong><\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"aligncenter size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"1024\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x.jpg\" alt=\"Et AI-behandlet bilde av en mann ved navn Ben, som holder et skilt med p\u00e5skriften "Abolish ICE".\" class=\"wp-image-5308\" style=\"width:537px;height:auto\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-300x300.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-150x150.jpg 150w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-768x768.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/Gemini_Generated_Image_qd4x5sqd4x5sqd4x-12x12.jpg 12w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n<\/div>\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 14%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banana<\/td>\n <td style=\"font-size:0.78rem;\">Deepfake \/ Disinfo<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">97.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">98.88% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">5.74% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Det tok 30 sekunder. Alt jeg trengte \u00e5 gj\u00f8re var \u00e5 dra og slippe hodeskuddet hans inn i Gemini og be den om \u00e5 \"holde opp et skilt der det st\u00e5r avskaff ICE\". Resultatet fra Nano Banana er ikke like fotorealistisk som de andre, men det ser likevel autentisk nok ut til \u00e5 v\u00e6re overbevisende.<\/p>\n\n\n\n<p style=\"font-size:17px\"><span style=\"box-sizing: border-box; margin: 0px; padding: 0px;\">Heldigvis identifiserte fire av detektorene bildet n\u00f8yaktig:\u00a0<strong>TruthScan<\/strong> (9<\/span>9% AI), Was It AI (99% AI), Sight Engine (97% AI) og AI or Not (97% AI). Winston AI mislyktes, og fikk bare 5% AI-poeng.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#9. \"Club Void\" (generert av Midjourney)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"771\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-1024x771.jpg\" alt=\"En smilende kvinne i en hvit, kortermet kjole st\u00e5r i et svakt opplyst, overfylt rom. Bildet ser ut til \u00e5 v\u00e6re tatt med blits, noe som skaper en disig, bl\u00e5lig atmosf\u00e6re med svake partikler synlige i luften. I bakgrunnen st\u00e5r andre, lettkledde mennesker under et tak som ser skadet eller industrielt ut.\" class=\"wp-image-5309\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-1024x771.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-300x226.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-768x578.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club-16x12.jpg 16w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-Girl-in-club.jpg 1232w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 14%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Midt p\u00e5 reisen <span style=\"font-size:0.75rem; color:#888; display:block;\">Full generasjon<\/span><\/td>\n <td>Generelt<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">89.21% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">1.56% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">70.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Jeg genererte bildet over ved hjelp av Midjourney. Jo mer jeg s\u00e5 p\u00e5 det, jo mer foruroligende f\u00f8ltes det. Estetikken er m\u00f8rk og uhyggelig. En skyggefull tilstedev\u00e6relse ser ut til \u00e5 dukke opp bak motivet. For \u00e5 skape dette bildet brukte jeg Midjourneys estetiske referansefunksjon og ba det ganske enkelt om \u00e5 v\u00e6re en \"jente som st\u00e5r i en klubb og smiler\". Jeg viste bildet til noen f\u00e5 personer, og alle trodde det var ekte. <br><br>De fleste detektorer flagget bildet som AI-generert. TruthScan og Sight Engine flagget det med en AI-poengsum p\u00e5 99%, mens AI or Not sa 89%. <\/p>\n\n\n\n<p style=\"font-size:17px\">WasItAI var ikke like sikker som detektorene, og klassifiserte bildet med bare 70% AI-poeng. <\/p>\n\n\n\n<p style=\"font-size:17px\">Winston AIs detektor mislyktes fullstendig, og ga bildet en AI-poengsum p\u00e5 1,56%. <\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#10 \"Ice Box\"<strong> <\/strong>(generert av Midjourney)<\/h2>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"771\" src=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-1024x771.jpg\" alt=\"Et uskarpt bilde som viser en mann med r\u00f8dflekket skjorte som st\u00e5r foran noen.\" class=\"wp-image-5311\" title=\"\" srcset=\"https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-1024x771.jpg 1024w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-300x226.jpg 300w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-768x578.jpg 768w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE-16x12.jpg 16w, https:\/\/research.undetectable.ai\/wp-content\/uploads\/2026\/02\/MJ-ICE.jpg 1232w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n .detector-scores-wrap {\n max-width: 780px;\n margin: 2rem auto;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n }\n\n .detector-scores-wrap h3 {\n font-family: 'Outfit', sans-serif;\n font-size: 1.4rem;\n font-weight: 800;\n color: #1a1a2e;\n margin-bottom: 1.25rem;\n letter-spacing: -0.02em;\n }\n\n .detector-scores-table {\n width: 100%;\n border-collapse: separate;\n border-spacing: 0;\n border-radius: 14px;\n overflow: hidden;\n table-layout: fixed;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n\n .detector-scores-table thead th {\n background: #1a1a2e;\n color: #ffffff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.7rem;\n text-transform: uppercase;\n letter-spacing: 0.04em;\n padding: 14px 6px;\n text-align: center;\n border: none;\n }\n\n .detector-scores-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 18%;\n }\n\n .detector-scores-table thead th:nth-child(2) {\n width: 14%;\n }\n\n .detector-scores-table tbody tr {\n transition: background 0.2s ease;\n }\n\n .detector-scores-table tbody tr:hover {\n background: rgba(0,0,0,0.02);\n }\n\n .detector-scores-table tbody td {\n padding: 14px 6px;\n text-align: center;\n font-size: 0.9rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #eee;\n }\n\n .detector-scores-table tbody tr:last-child td {\n border-bottom: none;\n }\n\n .detector-scores-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n }\n\n .score-pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.78rem;\n border-radius: 6px;\n padding: 5px 8px;\n display: inline-block;\n min-width: 0;\n text-align: center;\n white-space: nowrap;\n }\n\n .score-pass {\n background: #22c55e;\n color: #fff;\n }\n\n .score-warn {\n background: #fb923c;\n color: #fff;\n }\n\n .score-fail {\n background: #ef4444;\n color: #fff;\n }\n\n @media (max-width: 600px) {\n .detector-scores-table thead th,\n .detector-scores-table tbody td {\n padding: 10px 4px;\n font-size: 0.7rem;\n }\n .score-pill {\n padding: 4px 5px;\n font-size: 0.68rem;\n }\n .detector-scores-wrap h3 {\n font-size: 1.15rem;\n }\n }\n<\/style>\n\n<div class=\"detector-scores-wrap\">\n <h3>Slik vurderte bildedetektorene det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bildemodell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Midt p\u00e5 reisen <span style=\"font-size:0.75rem; color:#888; display:block;\">Full generasjon<\/span><\/td>\n <td style=\"font-size:0.82rem;\">Disinfo<\/td>\n <td><span class=\"score-pill score-pass\">99.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">92.00% AI<\/span><\/td>\n <td><span class=\"score-pill score-pass\">97.94% AI<\/span><\/td>\n <td><span class=\"score-pill score-fail\">41.20% AI<\/span><\/td>\n <td><span class=\"score-pill score-warn\">72.00% AI<\/span><\/td>\n <\/tr>\n <\/tbody>\n <\/table>\n<\/div>\n\n\n\n<p style=\"font-size:17px\">Dette bildet ble til ved hjelp av samme metode som i forrige test. Denne gangen var m\u00e5let \u00e5 generere noe litt m\u00f8rkere. TruthScan var den mest n\u00f8yaktige detektoren, og merket bildet som 99% AI. Sight Engine og AI or Not gjenkjente begge bildet korrekt. Winston AI mislyktes fullstendig, og WasItAI hadde en mye lavere tillitsscore enn TruthScan, Sight Engine og AI or Not.<\/p>\n\n\n\n<p style=\"font-size:17px\">Jeg m\u00e5 si at i dette tilfellet kan bildet jeg genererte brukes eller innrammes p\u00e5 to m\u00e5ter. P\u00e5 den ene siden kan noen bruke dette (eller andre medier med politisk\/borgerlige uroligheter som tema) i et slags kunstprosjekt eller kreativt prosjekt for \u00e5 komme med en uttalelse. I et slikt tilfelle virker risikoen lavere. Den st\u00f8rste bekymringen jeg har med denne kategorien AI-genererte medier, er at de blir brukt av skumle personer som hevder at de er ekte.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Oppsummering og resultater: Beste AI-bildedetektorer<\/h2>\n\n\n\n<p style=\"font-size:17px\">N\u00e5 som vi har vist alle testene, skal jeg oppsummere: Jeg genererte 10 AI-bilder<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li style=\"font-size:17px\">Jeg brukte ChatGPT, Nano Banana og Midjourney til \u00e5 generere 10 AI-bilder<\/li>\n\n\n\n<li style=\"font-size:17px\">Jeg testet fem AI-bildedetektorer ved \u00e5 sende alle AI-bildene jeg genererte, gjennom dem.<\/li>\n\n\n\n<li style=\"font-size:17px\">TruthScan besto alle testene, og var den mest n\u00f8yaktige detektoren. AI or Not besto 8 av 10 tester, og viste en viss p\u00e5litelighet. Sight Engine besto 3 av 10 tester, og hadde gjennomg\u00e5ende tvilsom n\u00f8yaktighet. Was It AI besto 4 av 10 tester, og hadde gjennomg\u00e5ende d\u00e5rlig n\u00f8yaktighet. Winston AI var den minst n\u00f8yaktige AI-bildedetektoren, og besto bare 3 av 10 tester, og feilklassifiserte konsekvent bilder.<\/li>\n<\/ul>\n\n\n\n<!DOCTYPE html>\n<html lang=\"en\">\n<head>\n<meta charset=\"UTF-8\">\n<meta name=\"viewport\" content=\"width=device-width, initial-scale=1.0\">\n<title>AI Image Detector - omfattende testresultater<\/title>\n<style>\n @import url('https:\/\/fonts.googleapis.com\/css2?family=DM+Sans:wght@400;500;600;700&family=Outfit:wght@600;700;800&display=swap');\n\n *,*::before,*::after{box-sizing:border-box;margin:0;padding:0}\n\n body {\n background: #f4f3f0;\n min-height: 100vh;\n font-family: 'DM Sans', sans-serif;\n -webkit-font-smoothing: antialiased;\n color: #2d2d3a;\n }\n\n \/* \u2500\u2500 wrapper \u2500\u2500 *\/\n .results-page {\n max-width: 860px;\n margin: 0 auto;\n padding: 3rem 1.25rem 4rem;\n }\n\n \/* \u2500\u2500 hero header \u2500\u2500 *\/\n .hero {\n text-align: center;\n margin-bottom: 2.75rem;\n }\n .hero-badge {\n display: inline-block;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.65rem;\n text-transform: uppercase;\n letter-spacing: 0.12em;\n color: #fff;\n background: #1a1a2e;\n padding: 6px 16px;\n border-radius: 100px;\n margin-bottom: 1.1rem;\n }\n .hero h1 {\n font-family: 'Outfit', sans-serif;\n font-weight: 800;\n font-size: 2.4rem;\n color: #1a1a2e;\n letter-spacing: -0.03em;\n line-height: 1.15;\n margin-bottom: 0.6rem;\n }\n .hero p {\n font-size: 1.02rem;\n color: #6b6b7b;\n max-width: 560px;\n margin: 0 auto;\n line-height: 1.55;\n }\n\n \/* \u2500\u2500 accuracy cards row \u2500\u2500 *\/\n .accuracy-cards {\n display: grid;\n grid-template-columns: repeat(5, 1fr);\n gap: 10px;\n margin-bottom: 2.75rem;\n }\n .acc-card {\n background: #fff;\n border-radius: 14px;\n padding: 1.15rem 0.6rem 1rem;\n text-align: center;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.05),\n 0 6px 20px rgba(0,0,0,0.07),\n 0 0 0 1px rgba(0,0,0,0.03);\n position: relative;\n overflow: hidden;\n transition: transform 0.2s ease, box-shadow 0.2s ease;\n }\n .acc-card:hover {\n transform: translateY(-3px);\n box-shadow:\n 0 2px 6px rgba(0,0,0,0.06),\n 0 12px 32px rgba(0,0,0,0.1),\n 0 0 0 1px rgba(0,0,0,0.04);\n }\n .acc-card::before {\n content: '';\n position: absolute;\n top: 0; left: 0; right: 0;\n height: 4px;\n }\n .acc-card.rank-1::before { background: #16a34a; }\n .acc-card.rank-2::before { background: #a3e635; }\n .acc-card.rank-3::before { background: #facc15; }\n .acc-card.rank-4::before { background: #fb923c; }\n .acc-card.rank-5::before { background: #ef4444; }\n\n .acc-card .rank-num {\n font-family: 'Outfit', sans-serif;\n font-weight: 800;\n font-size: 0.62rem;\n text-transform: uppercase;\n letter-spacing: 0.08em;\n color: #999;\n margin-bottom: 0.35rem;\n }\n .acc-card .tool-name {\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.92rem;\n color: #1a1a2e;\n margin-bottom: 0.65rem;\n line-height: 1.2;\n }\n .acc-ring {\n width: 72px; height: 72px;\n margin: 0 auto 0.55rem;\n position: relative;\n }\n .acc-ring svg { transform: rotate(-90deg); }\n .acc-ring .ring-bg { fill: none; stroke: #eee; stroke-width: 5; }\n .acc-ring .ring-fg { fill: none; stroke-width: 5; stroke-linecap: round; transition: stroke-dashoffset 1s ease; }\n .acc-ring .ring-label {\n position: absolute;\n inset: 0;\n display: flex;\n align-items: center;\n justify-content: center;\n font-family: 'Outfit', sans-serif;\n font-weight: 800;\n font-size: 1.1rem;\n color: #1a1a2e;\n }\n .acc-card .acc-fraction {\n font-size: 0.78rem;\n color: #888;\n font-weight: 500;\n }\n\n \/* \u2500\u2500 section titles \u2500\u2500 *\/\n .section-title {\n font-family: 'Outfit', sans-serif;\n font-weight: 800;\n font-size: 1.45rem;\n color: #1a1a2e;\n letter-spacing: -0.02em;\n margin-bottom: 1rem;\n }\n .section-subtitle {\n font-size: 0.88rem;\n color: #888;\n margin-bottom: 1.25rem;\n line-height: 1.5;\n }\n\n \/* \u2500\u2500 master table \u2500\u2500 *\/\n .master-table-wrap {\n border-radius: 14px;\n overflow: hidden;\n box-shadow:\n 0 1px 3px rgba(0,0,0,0.06),\n 0 8px 24px rgba(0,0,0,0.08),\n 0 0 0 1px rgba(0,0,0,0.04);\n margin-bottom: 2.75rem;\n background: #fff;\n }\n .master-table {\n width: 100%;\n border-collapse: collapse;\n table-layout: fixed;\n }\n .master-table thead th {\n background: #1a1a2e;\n color: #fff;\n font-family: 'Outfit', sans-serif;\n font-weight: 700;\n font-size: 0.62rem;\n text-transform: uppercase;\n letter-spacing: 0.05em;\n padding: 13px 5px;\n text-align: center;\n border: none;\n position: sticky;\n top: 0;\n z-index: 2;\n }\n .master-table thead th:first-child {\n text-align: left;\n padding-left: 14px;\n width: 14%;\n }\n .master-table thead th:nth-child(2) { width: 13%; }\n .master-table thead th:nth-child(3) { width: 9%; }\n\n .master-table tbody tr {\n transition: background 0.18s ease;\n }\n .master-table tbody tr:hover {\n background: rgba(26,26,46,0.025);\n }\n .master-table tbody td {\n padding: 11px 5px;\n text-align: center;\n font-size: 0.82rem;\n font-weight: 500;\n color: #2d2d3a;\n border-bottom: 1px solid #f0f0ec;\n }\n .master-table tbody tr:last-child td { border-bottom: none; }\n .master-table tbody td:first-child {\n text-align: left;\n padding-left: 14px;\n font-weight: 600;\n color: #1a1a2e;\n font-size: 0.78rem;\n line-height: 1.3;\n }\n .test-num {\n display: inline-block;\n font-family: 'Outfit', sans-serif;\n font-weight: 800;\n font-size: 0.6rem;\n background: #1a1a2e;\n color: #fff;\n width: 18px; height: 18px;\n line-height: 18px;\n text-align: center;\n border-radius: 5px;\n margin-right: 5px;\n vertical-align: middle;\n }\n .img-model-sub {\n display: block;\n font-size: 0.68rem;\n color: #999;\n font-weight: 400;\n margin-top: 1px;\n }\n .cat-badge {\n font-family: 'Outfit', sans-serif;\n font-size: 0.6rem;\n font-weight: 700;\n text-transform: uppercase;\n letter-spacing: 0.03em;\n padding: 3px 6px;\n border-radius: 5px;\n display: inline-block;\n white-space: normal;\n line-height: 1.35;\n max-width: 100%;\n }\n .cat-fraud { background: #fef2f2; color: #dc2626; }\n .cat-disinfo { background: #fefce8; color: #a16207; }\n .cat-general { background: #f0fdf4; color: #16a34a; }\n .cat-deepfake { background: #faf5ff; color: #7c3aed; }\n .cat-mixed { background: #fff7ed; color: #c2410c; }\n\n \/* score pills *\/\n .pill {\n font-weight: 700;\n font-family: 'Outfit', sans-serif;\n font-size: 0.7rem;\n border-radius: 6px;\n padding: 4px 7px;\n display: inline-block;\n text-align: center;\n white-space: nowrap;\n min-width: 68px;\n }\n .pill-pass { background: #22c55e; color: #fff; }\n .pill-warn { background: #fb923c; color: #fff; }\n .pill-fail { background: #ef4444; color: #fff; }\n\n \/* \u2500\u2500 legend \u2500\u2500 *\/\n .legend {\n display: flex;\n gap: 1.25rem;\n align-items: center;\n margin-bottom: 1.25rem;\n flex-wrap: wrap;\n }\n .legend-item {\n display: flex;\n align-items: center;\n gap: 6px;\n font-size: 0.78rem;\n font-weight: 500;\n color: #666;\n }\n .legend-dot {\n width: 12px; height: 12px;\n border-radius: 4px;\n }\n .legend-dot.lg-pass { background: #22c55e; }\n .legend-dot.lg-warn { background: #fb923c; }\n .legend-dot.lg-fail { background: #ef4444; }\n\n \/* \u2500\u2500 bottom summary strip \u2500\u2500 *\/\n .summary-strip {\n display: grid;\n grid-template-columns: repeat(3, 1fr);\n gap: 10px;\n margin-bottom: 2.5rem;\n }\n .strip-card {\n background: #1a1a2e;\n border-radius: 12px;\n padding: 1.1rem 1rem;\n color: #fff;\n text-align: center;\n }\n .strip-card .strip-val {\n font-family: 'Outfit', sans-serif;\n font-weight: 800;\n font-size: 1.65rem;\n letter-spacing: -0.02em;\n margin-bottom: 0.15rem;\n }\n .strip-card .strip-label {\n font-size: 0.72rem;\n color: rgba(255,255,255,0.55);\n font-weight: 500;\n text-transform: uppercase;\n letter-spacing: 0.05em;\n }\n .strip-val .hl-green { color: #4ade80; }\n .strip-val .hl-red { color: #f87171; }\n .strip-val .hl-amber { color: #fbbf24; }\n\n \/* \u2500\u2500 footer note \u2500\u2500 *\/\n .foot-note {\n font-size: 0.78rem;\n color: #999;\n line-height: 1.6;\n text-align: center;\n max-width: 640px;\n margin: 0 auto;\n }\n .foot-note strong { color: #666; font-weight: 600; }\n\n \/* \u2500\u2500 responsive \u2500\u2500 *\/\n @media (max-width: 700px) {\n .hero h1 { font-size: 1.6rem; }\n .accuracy-cards { grid-template-columns: repeat(2, 1fr); }\n .accuracy-cards .acc-card:last-child { grid-column: span 2; justify-self: center; max-width: 200px; }\n .summary-strip { grid-template-columns: 1fr; }\n .master-table thead th,\n .master-table tbody td { padding: 9px 3px; font-size: 0.68rem; }\n .pill { font-size: 0.6rem; padding: 3px 4px; min-width: 54px; }\n .test-num { width: 15px; height: 15px; line-height: 15px; font-size: 0.52rem; }\n .section-title { font-size: 1.15rem; }\n }\n\n \/* \u2500\u2500 entrance animations \u2500\u2500 *\/\n @keyframes fadeUp {\n from { opacity: 0; transform: translateY(18px); }\n to { opacity: 1; transform: translateY(0); }\n }\n .anim { animation: fadeUp 0.6s ease both; }\n .anim-d1 { animation-delay: 0.08s; }\n .anim-d2 { animation-delay: 0.16s; }\n .anim-d3 { animation-delay: 0.24s; }\n .anim-d4 { animation-delay: 0.32s; }\n .anim-d5 { animation-delay: 0.4s; }\n<\/style>\n<\/head>\n<body>\n\n<div class=\"results-page\">\n\n <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 HERO \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n <header class=\"hero anim\">\n <span class=\"hero-badge\">50 deteksjonssjekker - 5 detektorer - 10 AI-bilder<\/span>\n <h1>Omfattende testresultater<\/h1>\n <p>Hvordan fem popul\u00e6re AI-bildedetektorer presterte p\u00e5 tvers av svindel, desinformasjon, deepfakes og generell fotografering.<\/p>\n <\/header>\n\n <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 ACCURACY RING CARDS \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n <div class=\"accuracy-cards anim anim-d1\">\n\n <!-- 1 \u2014 TruthScan 100% -->\n <div class=\"acc-card rank-1\">\n <div class=\"rank-num\">#1<\/div>\n <div class=\"tool-name\">TruthScan<\/div>\n <div class=\"acc-ring\">\n <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n stroke=\"#16a34a\"\n stroke-dasharray=\"213.6\"\n stroke-dashoffset=\"0\"\/>\n <\/svg>\n <span class=\"ring-label\">100%<\/span>\n <\/div>\n <div class=\"acc-fraction\">10 \/ 10 oppdaget<\/div>\n <\/div>\n\n <!-- 2 \u2014 AI or Not 80% -->\n <div class=\"acc-card rank-2\">\n <div class=\"rank-num\">#2<\/div>\n <div class=\"tool-name\">AI eller ikke<\/div>\n <div class=\"acc-ring\">\n <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n stroke=\"#a3e635\"\n stroke-dasharray=\"213.6\"\n stroke-dashoffset=\"42.72\"\/>\n <\/svg>\n <span class=\"ring-label\">80%<\/span>\n <\/div>\n <div class=\"acc-fraction\">8 \/ 10 oppdaget<\/div>\n <\/div>\n\n <!-- 3 \u2014 Sight Engine 70% -->\n <div class=\"acc-card rank-3\">\n <div class=\"rank-num\">#3<\/div>\n <div class=\"tool-name\">Siktmotor<\/div>\n <div class=\"acc-ring\">\n <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n stroke=\"#facc15\"\n stroke-dasharray=\"213.6\"\n stroke-dashoffset=\"64.08\"\/>\n <\/svg>\n <span class=\"ring-label\">70%<\/span>\n <\/div>\n <div class=\"acc-fraction\">7 \/ 10 oppdaget<\/div>\n <\/div>\n\n <!-- 4 \u2014 WasItAI 60% -->\n <div class=\"acc-card rank-4\">\n <div class=\"rank-num\">#4<\/div>\n <div class=\"tool-name\">WasItAI<\/div>\n <div class=\"acc-ring\">\n <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n stroke=\"#fb923c\"\n stroke-dasharray=\"213.6\"\n stroke-dashoffset=\"85.44\"\/>\n <\/svg>\n <span class=\"ring-label\">60%<\/span>\n <\/div>\n <div class=\"acc-fraction\">6 \/ 10 oppdaget<\/div>\n <\/div>\n\n <!-- 5 \u2014 Winston AI 30% -->\n <div class=\"acc-card rank-5\">\n <div class=\"rank-num\">#5<\/div>\n <div class=\"tool-name\">Winston AI<\/div>\n <div class=\"acc-ring\">\n <svg viewbox=\"0 0 80 80\" width=\"72\" height=\"72\">\n <circle class=\"ring-bg\" cx=\"40\" cy=\"40\" r=\"34\"\/>\n <circle class=\"ring-fg\" cx=\"40\" cy=\"40\" r=\"34\"\n stroke=\"#ef4444\"\n stroke-dasharray=\"213.6\"\n stroke-dashoffset=\"149.52\"\/>\n <\/svg>\n <span class=\"ring-label\">30%<\/span>\n <\/div>\n <div class=\"acc-fraction\">3 \/ 10 oppdaget<\/div>\n <\/div>\n\n <\/div>\n\n <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 STATS STRIP \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n <div class=\"summary-strip anim anim-d2\">\n <div class=\"strip-card\">\n <div class=\"strip-val\"><span class=\"hl-green\">34<\/span> \/ 50<\/div>\n <div class=\"strip-label\">Totalt antall korrekte deteksjoner<\/div>\n <\/div>\n <div class=\"strip-card\">\n <div class=\"strip-val\"><span class=\"hl-red\">16<\/span> \/ 50<\/div>\n <div class=\"strip-label\">Totalt antall manglende oppdagelser<\/div>\n <\/div>\n <div class=\"strip-card\">\n <div class=\"strip-val\"><span class=\"hl-amber\">68%<\/span><\/div>\n <div class=\"strip-label\">Gjennomsnittlig n\u00f8yaktighet p\u00e5 tvers av alle<\/div>\n <\/div>\n <\/div>\n\n <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 LEGEND \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n <h3 class=\"section-title anim anim-d3\">Alle 10 testene - fullstendig oversikt<\/h3>\n <p class=\"section-subtitle anim anim-d3\">Hver poengsum representerer detektorens AI-konfidensvurdering. Terskelverdi for best\u00e5tt: \u2265 90% AI.<\/p>\n\n <div class=\"legend anim anim-d3\">\n <div class=\"legend-item\"><span class=\"legend-dot lg-pass\"><\/span> Pass (\u2265 90%)<\/div>\n <div class=\"legend-item\"><span class=\"legend-dot lg-warn\"><\/span> Nesten-uhell (70-89%)<\/div>\n <div class=\"legend-item\"><span class=\"legend-dot lg-fail\"><\/span> Feil (< 70%)<\/div>\n <\/div>\n\n <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 MASTER TABLE \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n <div class=\"master-table-wrap anim anim-d4\">\n <table class=\"master-table\">\n <thead>\n <tr>\n <th>Test<\/th>\n <th>Kategori<\/th>\n <th>Kilde<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller ikke<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n\n <!-- #1 -->\n <tr>\n <td><span class=\"test-num\">1<\/span> Mannen p\u00e5 kanten<\/td>\n <td><span class=\"cat-badge cat-general\">Generelt<\/span><\/td>\n <td><span class=\"img-model-sub\">ChatGPT<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-warn\">78.00%<\/span><\/td>\n <td><span class=\"pill pill-fail\">0.98%<\/span><\/td>\n <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n <\/tr>\n\n <!-- #2 -->\n <tr>\n <td><span class=\"test-num\">2<\/span> Den falske kvitteringen<\/td>\n <td><span class=\"cat-badge cat-fraud\">Bedrageri<\/span><\/td>\n <td><span class=\"img-model-sub\">ChatGPT<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-fail\">19.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">94.48%<\/span><\/td>\n <td><span class=\"pill pill-fail\">0.04%<\/span><\/td>\n <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n <\/tr>\n\n <!-- #3 -->\n <tr>\n <td><span class=\"test-num\">3<\/span> Farepakke<\/td>\n <td><span class=\"cat-badge cat-fraud\">Bedrageri<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banana<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">98.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.10%<\/span><\/td>\n <td><span class=\"pill pill-fail\">31.98%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <\/tr>\n\n <!-- #4 -->\n <tr>\n <td><span class=\"test-num\">4<\/span> Kakerlakkm\u00e5ltid<\/td>\n <td><span class=\"cat-badge cat-fraud\">Bedrageri<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banana<\/span><\/td>\n <td><span class=\"pill pill-pass\">97.15%<\/span><\/td>\n <td><span class=\"pill pill-fail\">18.00%<\/span><\/td>\n <td><span class=\"pill pill-warn\">85.87%<\/span><\/td>\n <td><span class=\"pill pill-warn\">72.29%<\/span><\/td>\n <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n <\/tr>\n\n <!-- #5 -->\n <tr>\n <td><span class=\"test-num\">5<\/span> Autovandalisme<\/td>\n <td><span class=\"cat-badge cat-mixed\">Bedrageri \/ desinformasjon<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banana<\/span><\/td>\n <td><span class=\"pill pill-pass\">97.48%<\/span><\/td>\n <td><span class=\"pill pill-fail\">18.00%<\/span><\/td>\n <td><span class=\"pill pill-warn\">89.52%<\/span><\/td>\n <td><span class=\"pill pill-fail\">0.02%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <\/tr>\n\n <!-- #6 -->\n <tr>\n <td><span class=\"test-num\">6<\/span> Soda Musk & Don<\/td>\n <td><span class=\"cat-badge cat-disinfo\">Disinfo<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banana<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-fail\">0.24%<\/span><\/td>\n <td><span class=\"pill pill-warn\">86.00%<\/span><\/td>\n <\/tr>\n\n <!-- #7 -->\n <tr>\n <td><span class=\"test-num\">7<\/span> Fortsatt i live<\/td>\n <td><span class=\"cat-badge cat-deepfake\">Deepfake<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banana<\/span><\/td>\n <td><span class=\"pill pill-pass\">97.49%<\/span><\/td>\n <td><span class=\"pill pill-pass\">95.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.27%<\/span><\/td>\n <td><span class=\"pill pill-fail\">0.15%<\/span><\/td>\n <td><span class=\"pill pill-fail\">1.00%<\/span><\/td>\n <\/tr>\n\n <!-- #8 -->\n <tr>\n <td><span class=\"test-num\">8<\/span> Misbruk av en Ben<\/td>\n <td><span class=\"cat-badge cat-mixed\">Deepfake \/ Disinfo<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banana<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">97.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">98.88%<\/span><\/td>\n <td><span class=\"pill pill-fail\">5.74%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <\/tr>\n\n <!-- #9 -->\n <tr>\n <td><span class=\"test-num\">9<\/span> Club Void<\/td>\n <td><span class=\"cat-badge cat-general\">Generelt<\/span><\/td>\n <td><span class=\"img-model-sub\">Midt p\u00e5 reisen<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-warn\">89.21%<\/span><\/td>\n <td><span class=\"pill pill-fail\">1.56%<\/span><\/td>\n <td><span class=\"pill pill-warn\">70.00%<\/span><\/td>\n <\/tr>\n\n <!-- #10 -->\n <tr>\n <td><span class=\"test-num\">10<\/span> Isboksen<\/td>\n <td><span class=\"cat-badge cat-disinfo\">Disinfo<\/span><\/td>\n <td><span class=\"img-model-sub\">Midt p\u00e5 reisen<\/span><\/td>\n <td><span class=\"pill pill-pass\">99.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">92.00%<\/span><\/td>\n <td><span class=\"pill pill-pass\">97.94%<\/span><\/td>\n <td><span class=\"pill pill-fail\">41.20%<\/span><\/td>\n <td><span class=\"pill pill-warn\">72.00%<\/span><\/td>\n <\/tr>\n\n <\/tbody>\n <\/table>\n <\/div>\n\n <!-- \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 FOOTNOTES \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550 -->\n <p class=\"foot-note anim anim-d5\">\n <strong>Metodikk:<\/strong> Hvert bilde ble sendt inn \u00e9n gang til hver detektor. En poengsum p\u00e5 <strong>\u2265 90% AI<\/strong> regnes som en korrekt deteksjon. Poeng mellom 70-89% er \"nesten feil\". Alt under 70% er en feil. AI or Not merket \u00e9n generasjon ved 78%, \u00e9n ved 85% og \u00e9n ved 89% - 89% behandles som en nesten-feil med en asterisk i oppsummeringstabellen.\n <\/p>\n\n<\/div>\n\n<\/body>\n<\/html>\n\n\n\n<h2 class=\"wp-block-heading\">Avsluttende bemerkninger og data<\/h2>\n\n\n\n<p style=\"font-size:17px\">Til sammen brukte jeg 25 timer p\u00e5 \u00e5 skrive denne artikkelen og gjennomf\u00f8re de grundige testene. Multimodale AI-deteksjonsverkt\u00f8y er fortsatt under utvikling, men det er tydelig at noen er mer n\u00f8yaktige enn andre. Etter \u00e5 ha testet alle verkt\u00f8yene, har TruthScan den mest n\u00f8yaktige <a href=\"https:\/\/truthscan.com\/ai-image-detector\" target=\"_blank\" rel=\"noopener\">AI-bildedetektor<\/a>. Testene taler for seg selv. <\/p>\n\n\n\n<p style=\"font-size:17px\">Hvis du vil ha tilgang til en CSV-kopi av dataene fra testingen jeg gjorde i denne artikkelen, kan du finne den <a href=\"https:\/\/docs.google.com\/spreadsheets\/d\/1iuapCicNcdG6dZCFd-dCmSRgUJ9GVCS2v9kh4PShRew\/edit?usp=sharing\" data-type=\"link\" data-id=\"https:\/\/docs.google.com\/spreadsheets\/d\/1iuapCicNcdG6dZCFd-dCmSRgUJ9GVCS2v9kh4PShRew\/edit?usp=sharing\" target=\"_blank\" rel=\"noopener\">her.<\/a> Regnearket jeg lenket til, inneholder alle de opprinnelige sp\u00f8rsm\u00e5lene og deteksjonsresultatene fra testene i denne artikkelen. <\/p>","protected":false},"excerpt":{"rendered":"<p>Du har sikkert sett flere AI-genererte bilder p\u00e5 nettet enn du er klar over. Noen ganger er det \u00e5penbart at et [...]<\/p>","protected":false},"author":15,"featured_media":5326,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"_themeisle_gutenberg_block_has_review":false,"footnotes":""},"categories":[34],"tags":[19,96,97],"class_list":["post-5287","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-news","tag-ai-detection","tag-ai-images","tag-detecting-deepfakes"],"_links":{"self":[{"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/posts\/5287","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/users\/15"}],"replies":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/comments?post=5287"}],"version-history":[{"count":27,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/posts\/5287\/revisions"}],"predecessor-version":[{"id":5334,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/posts\/5287\/revisions\/5334"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/media\/5326"}],"wp:attachment":[{"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/media?parent=5287"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/categories?post=5287"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/research.undetectable.ai\/nb\/wp-json\/wp\/v2\/tags?post=5287"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}