{"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\/sv\/jag-testade-5-ai-bilddetektorer-dessa-ar-de-basta-2026\/","title":{"rendered":"Jag testade 5 AI-bilddetektorer. Dessa \u00e4r de b\u00e4sta \u00e5r 2026:"},"content":{"rendered":"

Du har f\u00f6rmodligen sett fler AI-genererade foton p\u00e5 n\u00e4tet \u00e4n du tror. <\/strong><\/p>\n\n\n\n

Ibland \u00e4r det uppenbart att en bild har genererats av AI, men det blir allt sv\u00e5rare att avg\u00f6ra i takt med att de generativa bild- och videoverktygen f\u00f6rb\u00e4ttras. Nya verktyg som Googles Nano Banana Pro<\/a>och uppdateringar av OpenAI:s ChatGPT-bildmodell<\/a> l\u00e5ter anv\u00e4ndare snabbt generera syntetiska bilder som speglar verkliga bilder. Tidigare forskning visade att 85% av amerikanerna s\u00e4ger deepfakes urholkar f\u00f6rtroendet p\u00e5 n\u00e4tet<\/a>.<\/p>\n\n\n\n

Flera verktyg g\u00f6r anspr\u00e5k p\u00e5 att uppt\u00e4cka AI-genererade bilder, men fungerar de? <\/strong><\/h2>\n\n\n\n

Jag k\u00f6rde 50 detektionskontroller \u00f6ver fem av de mest popul\u00e4ra AI-bilddetektorerna och dokumenterade resultaten. Jag kommer inte bara att presentera all data i den h\u00e4r artikeln och f\u00f6rklara den, jag kommer ocks\u00e5 att l\u00e4nka till dokumentationen i slutet.<\/p>\n\n\n\n

De fem detektorer som jag anv\u00e4nde f\u00f6r dessa tester var TruthScan<\/a>, AI eller inte<\/a>, Siktmotor<\/a>, WasItAI<\/a>, och Winston AI<\/a>. Jag testade dessa verktyg genom att l\u00e5ta var och en analysera tv\u00e5 ChatGPT-genererade bilder, sex Nano Banana-genererade bilder och tv\u00e5 bilder genererade av Midjourney. Jag anv\u00e4nde flera olika promptstilar och tekniker (som jag kommer att g\u00e5 in p\u00e5 i detalj senare n\u00e4r jag bryter ner varje analyserad bild).<\/p>\n\n\n\n

Alla fem detektorerna fick i uppgift att uppt\u00e4cka AI-skapade bilder i kategorier som bedr\u00e4geri, desinformation, allm\u00e4n fotografering och deepfakes. F\u00f6ga f\u00f6rv\u00e5nande presterade inte alla detektorer bra. TruthScan var den enda detektorn som konsekvent klassificerade allt inneh\u00e5ll jag skickade in p\u00e5 97% eller h\u00f6gre.<\/strong><\/p>\n\n\n\n\n\n\n\n\nTestresultat f\u00f6r 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 Testresultat f\u00f6r AI Image Detector<\/h2>\n\n <table class=\"ai-detector-table\">\n <thead>\n <tr>\n <th>Verktyg f\u00f6r detektering<\/th>\n <th>Totalt antal tester<\/th>\n <th>Korrekt detekterad<\/th>\n <th>Missade (misslyckas)<\/th>\n <th>Noggrannhet<\/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 inte<\/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 eller inte m\u00e4rkte en AI-generation som 78%, en som 85% och en som 89%. Eftersom testet kr\u00e4ver minst 90% r\u00e4knar vi de tv\u00e5 f\u00f6rsta som misslyckade men behandlar 89% som en n\u00e4stan miss, d\u00e4rav asterisken...<\/p>\n<\/div>\n\n<\/body>\n<\/html>\n\n\n\n<p style=\"font-size:17px\">AI eller inte presterade ocks\u00e5 ganska bra. I sj\u00e4lva verket, medan AI eller inte klassificerade flera AI-genererade objekt under 90%-s\u00e4kerhet, var det den n\u00e4st mest exakta och konsekventa detektorn under mina tester, bakom TruthScan. \u00c4ven d\u00e4r AI eller inte misslyckades var dessa misslyckanden inte katastrofala, men det finns fortfarande utrymme f\u00f6r f\u00f6rb\u00e4ttringar. <\/p>\n\n\n\n<p style=\"font-size:17px\">Resten av de AI-bilddetektorer jag f\u00f6rs\u00f6kte presterade mycket s\u00e4mre. Sight Engine felklassificerade till exempel 3 bedr\u00e4gerirelaterade AI-bilder som autentiska.<\/p>\n\n\n\n<p style=\"font-size:17px\">Nu ska jag visa dig varje bild (av de 10 jag genererade), f\u00f6rklara hur jag skapade den och visa hur varje modell po\u00e4ngsatte den.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#1. \"Man on a Ledge\" (genereras 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>ChattGPT <span style=\"font-size:0.75rem; color:#888; display:block;\">Fullst\u00e4ndig generation<\/span><\/td>\n <td>Allm\u00e4nt<\/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\">F\u00f6r mig (och jag ser mycket AI-genererat inneh\u00e5ll varje dag) ser bilden ganska \u00f6vertygande ut vid f\u00f6rsta anblicken. Om du klickade p\u00e5 n\u00e5gons sociala medieprofil, bl\u00e4ddrade igenom deras fl\u00f6de och s\u00e5g den h\u00e4r bilden, skulle den omedelbart sticka ut som en AI-genererad falsk? F\u00f6r att vara r\u00e4ttvis f\u00f6rs\u00f6kte jag bli kreativ med prompten. H\u00e4r \u00e4r den jag anv\u00e4nde f\u00f6r att generera den h\u00e4r bilden:<\/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>Generera mig ett snapshot-kamerastilfoto av en man som st\u00e5r p\u00e5 toppen av ett tak och tittar ner p\u00e5 gatan, nattetid, han \u00e4r upplyst av kamerablixten, det finns ljussp\u00e5r fr\u00e5n bilarna nedanf\u00f6r, en kvinna st\u00e5r bakom honom med handen p\u00e5 l\u00e4pparna och ler, fotografiets estetik \u00e4r som ett foto som togs p\u00e5 en snapshot-kamera 2009 av collegev\u00e4nner som roade sig.<\/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\">Oavsett hur kreativ jag blev eller hur m\u00e5nga detaljer jag inkluderade i min uppmaning klassificerade TruthScan och Sight Engine fortfarande korrekt utdata som AI-genererad. AI eller inte-detektorn kom n\u00e4ra, men missade \u00e4nd\u00e5 m\u00e5let. <\/p>\n\n\n\n<p style=\"font-size:17px\">Winston och WasItAI hade helt fel n\u00e4r de klassificerade ChatGPT-bilden som verklig.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#2. \"Det falska kvittot\" (genererat 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=\"En bild som visar vad som verkar vara ett f\u00e4rgat kvitto\" 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>ChattGPT <span style=\"font-size:0.75rem; color:#888; display:block;\">Fullst\u00e4ndig generation<\/span><\/td>\n <td>Bedr\u00e4geri<\/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\">Jag ville generera ett falskt kvitto med ChatGPT f\u00f6r att se hur realistiskt det skulle se ut. Du kanske m\u00e4rker att bilden saknar en legitim adress, men strukturen ser fortfarande lite realistisk ut. Jag skulle s\u00e4ga att den h\u00e4r bilden \u00e4r mindre \u00f6vertygande \u00e4n den f\u00f6rsta jag hade ChatGPT generera, men de flesta av detektorerna tankade fortfarande n\u00e4r de analyserade den.<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li style=\"font-size:17px\">Av alla kontroller jag gjorde p\u00e5 den h\u00e4r bilden, <strong>TruthScan<\/strong> var den mest exakta. <\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>Siktmotor<\/strong> misslyckades verkligen h\u00e4r och visade d\u00e5ligt omd\u00f6me om en dokumentrelaterad AI-bild.<\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>AI eller inte<\/strong> presterade m\u00e4rkbart b\u00e4ttre i denna andra analys.<\/li>\n\n\n\n<li style=\"font-size:17px\"><strong>Winston AI <\/strong>presterade s\u00e4mst, och WasItAI misslyckades ocks\u00e5 helt.<\/li>\n<\/ul>\n\n\n\n<p style=\"font-size:17px\">Med tanke p\u00e5 att den h\u00e4r bilden \u00e4r associerad med bedr\u00e4gerikategorin \u00e4r det oroande att de flesta detektorerna felklassificerade den. <\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Den prompt jag anv\u00e4nde f\u00f6r att generera bilden<\/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>Generera en bild av ett kvitto fr\u00e5n Best Buy d\u00e4r det st\u00e5r att det totala beloppet \u00e4r en miljard dollar och att det finns en kaffefl\u00e4ck p\u00e5 det<\/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\">Det \u00e4r uppenbart att jag skrev en snabb och enkel prompt. \u00c4nd\u00e5 \u00e4r kvaliteten p\u00e5 ChatGPT-utdata visuellt \u00f6vertygande f\u00f6r blotta \u00f6gat. Inte s\u00e5 \u00f6vertygande f\u00f6r tv\u00e5 av de fem detektorer jag testade, verkar det som.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#3: \"Hazard Package\" (genereras med Nano Banana)<\/h2>\n\n\n\n<p><strong><em>Ist\u00e4llet f\u00f6r att bara l\u00e5ta Nano Banana generera en originalbild gav jag den en bild att \u00e4ndra. <\/em><\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">Tanken var att visa hur n\u00e5gon enkelt kan anv\u00e4nda AI f\u00f6r att skapa falska bevis och h\u00e4vda att de f\u00e5tt ett skadat\/farligt paket. F\u00f6rst tog jag ett foto med en iPhone 15 av ett tomt Amazon-paket som jag hittade.<\/p>\n\n\n\n<p><strong>H\u00e4r \u00e4r fotot jag tog:<\/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=\"Ett foto av ett Amazon-paket\" 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\">Sedan laddade jag upp fotot jag tog till Nano Banana och gav den f\u00f6ljande uppmaning:<\/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>Jag vill att du redigerar det h\u00e4r fotot, \u00e4ndra ingenting f\u00f6rutom d\u00e4r etiketten sitter, l\u00e4gg till skador p\u00e5 den och l\u00e4gg till svarta slamfl\u00e4ckar p\u00e5 l\u00e5dan som ser ut som riktiga fettfl\u00e4ckar<\/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-utg\u00e5ng:<\/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=\"Ett foto som visar ett paket\" 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banan <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n <td>Bedr\u00e4geri<\/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\">Den oljeliknande s\u00f6rjan som Nano Banana genererade i utmatningen hade en \u00f6vertygande smutsig glans, komplett med en fuktig bl\u00f6tl\u00e4ggningseffekt - inte alls tecknad. Lyckligtvis m\u00e4rkte de flesta av de detektorer jag k\u00f6rde den igenom korrekt bilden som en falsk. Den enda avvikaren var Winston, som \u00e5terigen presterade d\u00e5ligt och felklassificerade bilden som autentisk. AI or Not, TruthScan och WasItAI var de mest exakta detektorerna f\u00f6r att uppt\u00e4cka den h\u00e4r bilden.<\/p>\n\n\n\n<p style=\"font-size:17px\">Det \u00e4r oroande hur enkelt det verkar vara att ladda upp en bild till en chatbot och f\u00e5 den helt f\u00f6r\u00e4ndrad. Jag kan t\u00e4nka mig att bedragare anv\u00e4nder generativa bildverktyg f\u00f6r att f\u00f6rs\u00f6ka g\u00f6ra bedr\u00e4gliga returanspr\u00e5k eller \u00e4gna sig \u00e5t marknadsbedr\u00e4gerier.<\/p>\n\n\n\n<p style=\"font-size:17px\">Om e-handelsplattformar bara kr\u00e4ver en bild som bevis f\u00f6r att inleda ett krav p\u00e5 skadat paket, utan n\u00e5got tillf\u00f6rlitligt s\u00e4tt att uppt\u00e4cka AI-manipulation, uts\u00e4tts de f\u00f6r en viktig attackvektor. Till exempel Amazons officiella <a href=\"https:\/\/sellercentral.amazon.com\/help\/hub\/reference\/external\/GYW2EV9FEMJ3JJGR?locale=en-US\" target=\"_blank\" rel=\"noopener\">returpolicy f\u00f6r varor<\/a> bokstavligen s\u00e4ger<strong>, “<em>Hazmat, inklusive brandfarliga v\u00e4tskor eller gaser, \u00e4r inte returvara<\/em>.”<\/strong><\/p>\n\n\n\n<p style=\"font-size:17px\">S\u00e5 om du faktiskt fick ett paket t\u00e4ckt av olja eller oljigt svart slam skulle du inte beh\u00f6va returnera det, men du kan fortfarande f\u00e5 en \u00e5terbetalning. Om det av n\u00e5gon anledning faktiskt h\u00e4nde dig skulle Amazon sannolikt be om fotobevis. F\u00f6rst\u00e5r du vilket problem AI-genererade bilder kan utg\u00f6ra n\u00e4r de inte g\u00e5r att skilja fr\u00e5n riktiga bilder?<\/p>\n\n\n\n<h2 class=\"wp-block-heading has-text-align-left\">#4. \"Cockroach Meal\" (genereras med Nano Banana)<\/h2>\n\n\n\n<p style=\"font-size:17px\">N\u00e4sta AI-bild kan f\u00e5 dig att tappa aptiten. Det visar sig att det finns dokumenterade fall d\u00e4r m\u00e4nniskor har anv\u00e4nt AI f\u00f6r att f\u00e5 <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\">bedr\u00e4gliga \u00e5terbetalningar<\/a> p\u00e5 appar f\u00f6r matleveranser. F\u00f6r detta fj\u00e4rde test f\u00f6ljde jag samma steg som i test tre. Den h\u00e4r g\u00e5ngen tog jag en bild av min tomma matl\u00e5da efter lunch.<\/p>\n\n\n\n<p><strong>H\u00e4r \u00e4r den riktiga bilden:<\/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=\"Ett foto av en tom matl\u00e5da f\u00f6r h\u00e4mtmat.\" 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\">Sedan lade jag in det riktiga fotot jag tog i Nano Banana och gav den f\u00f6ljande uppmaning:<\/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>\"Redigera den h\u00e4r bilden. \u00c4ndra ingenting f\u00f6rutom att l\u00e4gga till nudlar d\u00e4r folien \u00e4r och l\u00e4gga till en massa sm\u00e5 sm\u00e5 babykackerlackor i maten. F\u00e5 allt att se realistiskt ut och inte tecknad.\" <\/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=\"Ett AI-\u00e4ndrat foto av en matl\u00e5da med nudlar och insekter inuti l\u00e5dan\" 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banan <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n <td>Bedr\u00e4geri<\/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\">\u00c5terigen uppt\u00e4ckte TruthScan med s\u00e4kerhet AI-genererad media. Sight Engine och WasitAI misslyckades helt. AI or Not var den n\u00e4st mest exakta detektorn (med 85% AI-s\u00e4kerhet) n\u00e4r den skannade den h\u00e4r AI-bilden, och Winston presterade b\u00e4ttre \u00e4n i det tredje testet (med 72% AI-s\u00e4kerhet).<\/p>\n\n\n\n<p style=\"font-size:17px\">I den h\u00e4r bilden genererades nudlarna och kackerlackorna helt och h\u00e5llet av AI, men tv\u00e5 av detektorerna verkade inte tro det och endast en (TruthScan) var \u00f6ver 90% s\u00e4ker p\u00e5 AI-manipulation.<\/p>\n\n\n\n<p style=\"font-size:17px\">Ett problem h\u00e4r var hur enkelt det var att skapa den h\u00e4r falska bilden. Fr\u00e5n att ladda upp fotot jag tog, till att uppmana Nano Banana och ta emot resultatet, tog det bara cirka 2 minuter. En mardr\u00f6m f\u00f6r matleveransappar; en dr\u00f6m f\u00f6r hungriga l\u00f6gnare. En \"bedr\u00e4gerifaktor\".<\/p>\n\n\n\n<p class=\"has-vivid-red-color has-text-color has-link-color wp-elements-4c89b6ee3afcbdd6b6617c150e5b2e03\"><strong>OFFENTLIGG\u00d6RANDE: N\u00e5gra av de AI-genererade bilder som ing\u00e5r i f\u00f6ljande tester var avsedda att visa hur AI-bilder kan anv\u00e4ndas f\u00f6r desinformation. De inneh\u00e5ller kontroversiella bilder och k\u00e4nsliga \u00e4mnen<\/strong>, <strong>och \u00e4r INTE avsedda att representera n\u00e5gon s\u00e4rskild politisk<\/strong> <strong>\u00e5sikter eller ideologier.<\/strong><\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>#5<\/strong>. \"Autovandalism\" (genereras med Nano Banana)<\/h2>\n\n\n\n<p style=\"font-size:17px\">Jag hittade en riktig bild av en bil som var till salu p\u00e5 n\u00e4tet. Ganska vardagligt. Men t\u00e4nk om n\u00e5gon ville anv\u00e4nda den h\u00e4r bilden f\u00f6r att sprida desinformation eller l\u00e4mna in ett falskt bilf\u00f6rs\u00e4kringsanspr\u00e5k? Tyv\u00e4rr g\u00f6r generativ AI det snabbt och enkelt.<\/p>\n\n\n\n<p style=\"font-size:17px\"><strong>I det femte testet tog jag en riktig bild av en bil och gav Nano Banana f\u00f6ljande uppmaning:<\/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>\u00c4ndra ingenting p\u00e5 bilen f\u00f6rutom att jag vill att du sprayar p\u00e5 sprayf\u00e4rg med texten \"Free Palestine\" och att vindrutan \u00e4r sprucken. Str\u00e5lkastarna \u00e4r krossade och spegeln \u00e4r trasig och h\u00e4nger ner och de tv\u00e5 f\u00f6nstren p\u00e5 sidan av bilen \u00e4r<\/em><\/strong> <strong>krossad<\/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>H\u00e4r \u00e4r den riktiga bilden:<\/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=\"Ett foto 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-bild:<\/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=\"Ett AI-\u00e4ndrat foto av den bl\u00e5 bilen, som visar en sprucken vindruta och spraym\u00e5lningsskador p\u00e5 sidan 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banan <span style=\"font-size:0.75rem; color:#888; display:block;\">Deepfake Edit<\/span><\/td>\n <td style=\"font-size:0.82rem;\">Bedr\u00e4geri \/ desinformation<\/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\">Detektorresultaten f\u00f6r denna utg\u00e5ng var blandade. F\u00f6r detta test hade WasItAI (\u00f6verraskande) den h\u00f6gsta AI-detekteringspo\u00e4ngen (vid 99%). TruthScan hade den n\u00e4st h\u00f6gsta po\u00e4ngen (97%). Sight Engine misslyckades avsev\u00e4rt med att uppt\u00e4cka n\u00e5gra syntetiska element i bilden, och Winston gjorde det s\u00e4msta av alla detektorer (tilldelade en mindre \u00e4n 1% chans att involvera). AI or Not klarade sig bra, men tr\u00e4ffs\u00e4kerheten var fortfarande under 90%.<br><br>L\u00e5t oss prata om bilden. De tv\u00e5 mest oroande aspekterna av AI-skapat eller \u00e4ndrat inneh\u00e5ll som detta \u00e4r att b\u00e5da involverar bedr\u00e4geri. F\u00f6r det f\u00f6rsta \u00e4r det snabbt, enkelt och ser \u00e4kta ut att anv\u00e4nda AI f\u00f6r att \u00e4ndra vardagliga bilder f\u00f6r att skapa politiskt laddat inneh\u00e5ll. Oavsett om de som skapar den h\u00e4r typen av inneh\u00e5ll bara siktar p\u00e5 engagemang eller \u00e4r politiska provokat\u00f6rer, \u00e4r problemet att det \u00e4r falsk information som presenteras som autentisk.<\/p>\n\n\n\n<p style=\"font-size:17px\">Det andra s\u00e4ttet att anv\u00e4nda bildgenereringsverktyg som detta kan missbrukas \u00e4r f\u00f6r bedr\u00e4geri. Om en bedragare vill l\u00e4mna in ett falskt bilf\u00f6rs\u00e4kringsanspr\u00e5k m\u00e5ste de producera falska bevis. Ist\u00e4llet f\u00f6r att spendera timmar i Photoshop p\u00e5 att redigera bilder och f\u00f6rfalska bevis kan de anv\u00e4nda bildgeneratorer f\u00f6r att snabbt producera falska bevis.<\/p>\n\n\n\n<p style=\"font-size:17px\">I exemplet med AI-bilden ovan ser du hur en bedragare kan anv\u00e4nda en liknande process f\u00f6r att l\u00e4mna in en falsk skadeanm\u00e4lan enligt 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\">Policy f\u00f6r omfattande t\u00e4ckning<\/a>, som s\u00e4rskilt omfattar:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Slashed eller <a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-tire-damage\/\" target=\"_blank\" rel=\"noopener\">skadade d\u00e4ck<\/a><\/strong><\/li>\n\n\n\n<li><strong>Trasiga rutor, str\u00e5lkastare eller baklyktor<\/strong><\/li>\n\n\n\n<li><strong>Skador p\u00e5 sprayf\u00e4rg<\/strong><\/li>\n\n\n\n<li><strong><a href=\"https:\/\/www.progressive.com\/answers\/does-car-insurance-cover-scratches\/\" target=\"_blank\" rel=\"noopener\">Bucklor eller repor<\/a> fr\u00e5n n\u00e5gon som stj\u00e4l din bil<\/strong><\/li>\n\n\n\n<li><strong>Fylla p\u00e5 socker eller andra \u00e4mnen i bensintanken<\/strong><\/li>\n<\/ul>\n\n\n\n<p>Att l\u00e4mna in falska bevis \u00e4r naturligtvis inte bara vilseledande, det \u00e4r ocks\u00e5 direkt olagligt. I en idealisk v\u00e4rld skulle ingen bryta mot lagen, men i verkligheten sker brott varje dag. Och utan tillf\u00f6rlitlig uppt\u00e4ckt riskerar bilf\u00f6rs\u00e4kringsbolagen att bli lurade p\u00e5 miljontals kronor i takt med att AI-aktiverade falska f\u00f6rs\u00e4kringsanspr\u00e5k \u00f6kar.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#6. \"Soda Musk and Don\" (genereras 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=\"En AI-genererad bild av Elon Musk och 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banan <span style=\"font-size:0.75rem; color:#888; display:block;\">Fullst\u00e4ndig generation<\/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\">\u00c5terigen anv\u00e4nde jag Nano Banana f\u00f6r det h\u00e4r testet, och m\u00e5let var att demonstrera falska, politiskt laddade bilder med en mer f\u00e5nig underton. TruthScan, Sight Engine och AIorNot flaggade alla den genererade bilden med ett 99% AI-detekteringsbetyg. WasitAI var inte lika sj\u00e4lvs\u00e4kert - detekterade den AI-genererade bilden med endast 86% s\u00e4kerhet. Och Winston misslyckades helt och tilldelade en AI-po\u00e4ng p\u00e5 0,2%. F\u00f6r att generera bilden gav jag Nano Banana den h\u00e4r prompten:<\/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>Skapa ett foto av Elon Musk som spiller en stor mugg l\u00e4sk som en Big Gulp p\u00e5 sin skjorta och flippar ut medan Donald Trump sitter bredvid honom och skrattar. De befinner sig p\u00e5 ett jetplan. Det \u00e4r scenen. Och fotot ska se realistiskt ut som om det togs med en snapshot-kamera och inneh\u00e5lla artefakter som kamerablixten som reflekteras p\u00e5 flygplansf\u00f6nstret och utanf\u00f6r flygplanet. Det \u00e4r nattetid. Det ska se hyperrealistiskt 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\u00e4r ett av dessa program misslyckas med att uppt\u00e4cka AI betecknar de det som \u00e4kta. Hur f\u00e5nigt det h\u00e4r bildexemplet fr\u00e5n testet \u00e4n m\u00e5 vara, finns det inget roligt med att tro att en deepfake \u00e4r \u00e4kta.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#7. \"Still Alive\" (genereras 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=\"En AI-genererad bild av en man som h\u00e5ller i en skylt med texten "hj\u00e4lp"\" 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banan <span style=\"font-size:0.75rem; color:#888; display:block;\">Fullst\u00e4ndig generation<\/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\">Tycker du att den h\u00e4r bilden \u00e4r st\u00f6rande? F\u00f6r att generera denna deepfake anv\u00e4nde jag Nano Banana och gav den en bild av mig sj\u00e4lv. D\u00e4refter gav jag den en l\u00e5ng prompt som skulle f\u00f6rs\u00e4tta mig i en situation som antydde att jag hade blivit bortf\u00f6rd. Kanske \u00e4r det f\u00f6r att det h\u00e4r fotot visar <em>mig<\/em> i en komprometterande position. T\u00e4nk om n\u00e5gon gjorde n\u00e5got s\u00e5dant h\u00e4r, skickade det till min mamma och hotade att skada mig om hon inte skickade pengar till dem? L\u00e4skiga grejer. <\/p>\n\n\n\n<p class=\"has-small-font-size\"><em>(Den fullst\u00e4ndiga uppmaningen som jag anv\u00e4nde f\u00f6r detta finns i den fullst\u00e4ndiga testdatasetet i slutet av den h\u00e4r artikeln).<\/em><\/p>\n\n\n\n<p style=\"font-size:17px\"><br>Lyckligtvis f\u00f6r den h\u00e4r bilden, trots att den har ett mycket realistiskt utseende, <strong>tre av fem detektorer identifierade den som AI-genererad.<\/strong> <\/p>\n\n\n\n<p style=\"font-size:17px\">AI or Not kom in med en stark 99% AI-po\u00e4ng, TruthScan med 97% och Sight Engine p\u00e5 95%. Tyv\u00e4rr klassificerade Winston och WasItAI bilden som verklig.<\/p>\n\n\n\n<p style=\"font-size:17px\">Jag b\u00f6r klarg\u00f6ra att dessa tester inte gjordes p\u00e5 mitt personliga Gemini-konto. Nano Banana bad mig aldrig att bevisa att jag var personen p\u00e5 bilden. Vem som helst kunde ha laddat ner ett foto av mig fr\u00e5n internet, lagt in det i Nano Banana och f\u00e5tt det att skapa den h\u00e4r typen av bild. <\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#8. \"Abuse of a Ben\" (genereras 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\u00f6t Ben. Han tillbringar det mesta av sin tid med att leda olika avdelningar p\u00e5 de f\u00f6retag han arbetar p\u00e5. Bilden ovan \u00e4r Bens (riktiga) snygga f\u00f6retagsfoto. Om du inte m\u00e4rker det s\u00e5 \u00e4r Ben en professionell kille. Oavsett vilka politiska \u00e5sikter han har h\u00e5ller han dem f\u00f6r sig sj\u00e4lv, och hans samtal p\u00e5 jobbet handlar bara om jobbet. Men n\u00e4r som helst kan Bens bild anv\u00e4ndas f\u00f6r att sprida politisk propaganda eller budskap utan hans medgivande.<\/p>\n\n\n\n<p style=\"font-size:17px\">Den senaste tiden har det rapporterats en hel del om hur AI-genererade videor med ICE-agenter och protester har spridits p\u00e5 n\u00e4tet. Baserat p\u00e5 alla tester hittills drog jag slutsatsen att det var v\u00e4ldigt enkelt att skapa falsk politisk propaganda med AI. Jag best\u00e4mde mig f\u00f6r att testa igen, den h\u00e4r g\u00e5ngen med Ben. <\/p>\n\n\n\n<p style=\"font-size:17px\">Med Bens till\u00e5telse anv\u00e4nde jag hans huvudbild f\u00f6r att skapa en deepfake. <strong>H\u00e4r \u00e4r den AI-genererade bilden som Nano Banana skapade:<\/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=\"Ett AI-\u00e4ndrat foto av en man vid namn Ben som h\u00e5ller i en skylt med texten 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>Nano Banan<\/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 tog trettio sekunder. Allt jag beh\u00f6vde g\u00f6ra var att dra och sl\u00e4ppa hans huvudbild i Gemini och uppmana den att \"h\u00e5lla upp en skylt som s\u00e4ger avskaffa ICE.\" Resultatet fr\u00e5n Nano Banana \u00e4r inte lika fotorealistiskt som de andra, men det ser \u00e4nd\u00e5 tillr\u00e4ckligt autentiskt ut f\u00f6r att vara \u00f6vertygande.<\/p>\n\n\n\n<p style=\"font-size:17px\"><span style=\"box-sizing: border-box; margin: 0px; padding: 0px;\">Lyckligtvis identifierade fyra av detektorerna bilden korrekt:\u00a0<strong>TruthScan<\/strong> (9<\/span>9% AI), Var det AI (99% AI), Sight Engine (97% AI) och AI eller inte (97% AI). Winston AI misslyckades och tilldelades endast 5% AI-po\u00e4ng.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#9. \"Club Void\" (genererat 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 leende kvinna i vit, kort\u00e4rmad kl\u00e4nning st\u00e5r i ett svagt upplyst, tr\u00e5ngt rum. Bilden verkar vara tagen med blixt, vilket skapar en disig, bl\u00e5aktig atmosf\u00e4r med svaga partiklar synliga i luften. Andra m\u00e4nniskor i vardagskl\u00e4der st\u00e5r i bakgrunden under ett skadat eller industriellt utseende tak.\" 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>MidJourney <span style=\"font-size:0.75rem; color:#888; display:block;\">Fullst\u00e4ndig generation<\/span><\/td>\n <td>Allm\u00e4nt<\/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\">Jag skapade bilden ovan med hj\u00e4lp av Midjourney. Ju mer jag tittade p\u00e5 den, desto mer orov\u00e4ckande k\u00e4ndes den. Estetiken \u00e4r m\u00f6rk och kuslig. En skugglik n\u00e4rvaro verkar dyka upp bakom motivet. F\u00f6r att skapa den h\u00e4r bilden anv\u00e4nde jag Midjourneys estetiska referensfunktion och bad den helt enkelt att vara en \"tjej som st\u00e5r p\u00e5 en klubb och ler\". Jag visade bilden f\u00f6r n\u00e5gra personer och de trodde alla att den var verklig. <br><br>De flesta detektorer flaggade bilden som AI-genererad. TruthScan och Sight Engine flaggade den med en AI-po\u00e4ng p\u00e5 99%, AI eller inte sa 89%. <\/p>\n\n\n\n<p style=\"font-size:17px\">WasItAI var inte lika s\u00e4ker som detektorerna och klassificerade bilden med endast en 70% AI-po\u00e4ng. <\/p>\n\n\n\n<p style=\"font-size:17px\">Winston AI:s detektor misslyckades fullst\u00e4ndigt och gav bilden en AI-po\u00e4ng p\u00e5 1,56%. <\/p>\n\n\n\n<h2 class=\"wp-block-heading\">#10 \"Isbox\"<strong> <\/strong>(genererad 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=\"Ett suddigt foto visar en man med en r\u00f6dfl\u00e4ckig skjorta som st\u00e5r framf\u00f6r n\u00e5gon.\" 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>Hur bilddetektorerna bed\u00f6mde det<\/h3>\n\n <table class=\"detector-scores-table\">\n <thead>\n <tr>\n <th>Bild Modell<\/th>\n <th>Kategori<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/th>\n <th>Winston AI<\/th>\n <th>WasItAI<\/th>\n <\/tr>\n <\/thead>\n <tbody>\n <tr>\n <td>MidJourney <span style=\"font-size:0.75rem; color:#888; display:block;\">Fullst\u00e4ndig generation<\/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\">Jag skapade den h\u00e4r bilden med hj\u00e4lp av samma metod mitt under resan som i det f\u00f6reg\u00e5ende testet. Den h\u00e4r g\u00e5ngen var m\u00e5let att generera n\u00e5got lite m\u00f6rkare. TruthScan var den mest exakta detektorn och m\u00e4rkte bilden som 99% AI. Sight Engine och AI or Not uppt\u00e4ckte b\u00e5da bilden korrekt. Winston AI misslyckades helt och WasItAI hade ett mycket l\u00e4gre konfidensv\u00e4rde \u00e4n TruthScan, Sight Engine och AI or Not.<\/p>\n\n\n\n<p style=\"font-size:17px\">Jag borde s\u00e4ga att i det h\u00e4r fallet kan den bild jag genererade anv\u00e4ndas eller inramas p\u00e5 tv\u00e5 s\u00e4tt. \u00c5 ena sidan kan n\u00e5gon anv\u00e4nda den h\u00e4r typen av media (eller n\u00e5gon annan typ av media med temat politiska\/civila oroligheter) i n\u00e5gon form av konst eller kreativt projekt f\u00f6r att g\u00f6ra ett uttalande. I ett s\u00e5dant fall verkar risken l\u00e4gre. Den st\u00f6rsta oro jag har med denna kategori av AI-genererade medier \u00e4r att den anv\u00e4nds av sk\u00e4ndliga individer som h\u00e4vdar att den \u00e4r verklig.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">Sammanfattning och resultat: B\u00e4sta AI-bilddetektorer<\/h2>\n\n\n\n<p style=\"font-size:17px\">Okej, s\u00e5 nu n\u00e4r vi har visat alla tester, ska jag sammanfatta: Jag genererade 10 AI-bilder<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li style=\"font-size:17px\">Jag anv\u00e4nde ChatGPT, Nano Banana och Midjourney f\u00f6r att generera 10 AI-bilder<\/li>\n\n\n\n<li style=\"font-size:17px\">Jag testade fem AI-bilddetektorer genom att l\u00e5ta alla AI-bilder som jag genererade g\u00e5 igenom dem.<\/li>\n\n\n\n<li style=\"font-size:17px\">TruthScan klarade alla tester och var den mest exakta detektorn. AI or Not klarade 8 av 10 tester och visade viss tillf\u00f6rlitlighet. Sight Engine misslyckades med 3 av 10 tester och visade tvivelaktig noggrannhet. Was It AI misslyckades i 4 av 10 tester och hade d\u00e5lig noggrannhet \u00f6verlag. Winston AI var den minst noggranna AI-bilddetektorn, klarade bara 3 av 10 tester och klassificerade konsekvent bilder fel.<\/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 - omfattande testresultat<\/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 detektionskontroller - 5 detektorer - 10 AI-bilder<\/span>\n <h1>Omfattande testresultat<\/h1>\n <p>Hur fem popul\u00e4ra AI-bilddetektorer presterade i fr\u00e5ga om bedr\u00e4geri, desinformation, deepfakes och allm\u00e4n 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 uppt\u00e4ckt<\/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 inte<\/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 uppt\u00e4ckt<\/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 uppt\u00e4ckt<\/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 uppt\u00e4ckt<\/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 uppt\u00e4ckt<\/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 antal korrekta detektioner<\/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 antal missade detektioner<\/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\">Genomsnittlig noggrannhet f\u00f6r alla<\/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\">Alla 10 tester - fullst\u00e4ndig uppdelning<\/h3>\n <p class=\"section-subtitle anim anim-d3\">Varje po\u00e4ng representerar detektorns AI-f\u00f6rtroendebetyg. Tr\u00f6skelv\u00e4rde f\u00f6r godk\u00e4nt: \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> N\u00e4ra-miss (70-89%)<\/div>\n <div class=\"legend-item\"><span class=\"legend-dot lg-fail\"><\/span> Fel (< 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>K\u00e4lla<\/th>\n <th>TruthScan<\/th>\n <th>Siktmotor<\/th>\n <th>AI eller inte<\/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> Man p\u00e5 en klippavsats<\/td>\n <td><span class=\"cat-badge cat-general\">Allm\u00e4nt<\/span><\/td>\n <td><span class=\"img-model-sub\">ChattGPT<\/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> Det falska kvittot<\/td>\n <td><span class=\"cat-badge cat-fraud\">Bedr\u00e4geri<\/span><\/td>\n <td><span class=\"img-model-sub\">ChattGPT<\/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> Riskpaket<\/td>\n <td><span class=\"cat-badge cat-fraud\">Bedr\u00e4geri<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banan<\/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> Kackerlacka M\u00e5ltid<\/td>\n <td><span class=\"cat-badge cat-fraud\">Bedr\u00e4geri<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banan<\/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> Autovandalism<\/td>\n <td><span class=\"cat-badge cat-mixed\">Bedr\u00e4geri \/ desinformation<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banan<\/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 Banan<\/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> Fortfarande vid liv<\/td>\n <td><span class=\"cat-badge cat-deepfake\">Deepfake<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banan<\/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> Misshandel av en Ben<\/td>\n <td><span class=\"cat-badge cat-mixed\">Deepfake \/ Disinfo<\/span><\/td>\n <td><span class=\"img-model-sub\">Nano Banan<\/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\">Allm\u00e4nt<\/span><\/td>\n <td><span class=\"img-model-sub\">MidJourney<\/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> Isboxen<\/td>\n <td><span class=\"cat-badge cat-disinfo\">Disinfo<\/span><\/td>\n <td><span class=\"img-model-sub\">MidJourney<\/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>Metodik:<\/strong> Varje bild skickades in en g\u00e5ng till varje detektor. En po\u00e4ng p\u00e5 <strong>\u2265 90% AI<\/strong> r\u00e4knas som en korrekt detektering. Po\u00e4ng mellan 70-89% \u00e4r \"n\u00e4ra missar\". Allt under 70% \u00e4r ett misslyckande. AI eller inte m\u00e4rkte en generation vid 78%, en vid 85% och en vid 89% - 89% behandlas som en n\u00e4ra-miss med en asterisk i sammanfattningstabellen.\n <\/p>\n\n<\/div>\n\n<\/body>\n<\/html>\n\n\n\n<h2 class=\"wp-block-heading\">Avslutande kommentarer och data<\/h2>\n\n\n\n<p style=\"font-size:17px\">Sammanlagt har jag lagt 25 timmar p\u00e5 att sammanst\u00e4lla den h\u00e4r rapporten, fr\u00e5n det att jag skrev den h\u00e4r artikeln till de rigor\u00f6sa testerna. Verktygen f\u00f6r multimodal AI-detektering \u00e4r fortfarande under utveckling, men det \u00e4r tydligt att vissa \u00e4r mer exakta \u00e4n andra. Efter att ha testat alla verktyg har TruthScan den mest exakta <a href=\"https:\/\/truthscan.com\/ai-image-detector\" target=\"_blank\" rel=\"noopener\">AI-bilddetektor<\/a>. Testerna talar f\u00f6r sig sj\u00e4lva. <\/p>\n\n\n\n<p style=\"font-size:17px\">Om du vill ha tillg\u00e5ng till en CSV-kopia av data fr\u00e5n de tester jag gjorde i den h\u00e4r artikeln kan du hitta 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\">h\u00e4r.<\/a> Det kalkylblad som jag l\u00e4nkade till inneh\u00e5ller alla de ursprungliga uppmaningarna och detektionsresultaten fr\u00e5n testerna i den h\u00e4r artikeln. <\/p>","protected":false},"excerpt":{"rendered":"<p>Du har f\u00f6rmodligen sett fler AI-genererade foton online \u00e4n du inser. Ibland \u00e4r det uppenbart att en [...]<\/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\/sv\/wp-json\/wp\/v2\/posts\/5287","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/users\/15"}],"replies":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/comments?post=5287"}],"version-history":[{"count":27,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/posts\/5287\/revisions"}],"predecessor-version":[{"id":5334,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/posts\/5287\/revisions\/5334"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/media\/5326"}],"wp:attachment":[{"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/media?parent=5287"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/categories?post=5287"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/research.undetectable.ai\/sv\/wp-json\/wp\/v2\/tags?post=5287"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}