{"id":114550,"date":"2026-02-13T23:58:13","date_gmt":"2026-02-13T23:58:13","guid":{"rendered":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/?p=114550"},"modified":"2026-05-01T12:36:07","modified_gmt":"2026-05-01T12:36:07","slug":"exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai","status":"publish","type":"post","link":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/","title":{"rendered":"TOPstory &#8211;\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\"><strong>Le spiegazioni controfattuali, pensate per rendere l\u2019AI pi\u00f9 trasparente, possono diventare un\u2019arma per ricostruire modelli riservati. Un nuovo studio teorico mostra come, nel caso di modelli lineari, bastino pochissime query per comprometterne la sicurezza.<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">L\u2019estrazione di modelli (<em>model extraction<\/em>) \u00e8 <a href=\"https:\/\/arxiv.org\/pdf\/2602.02027\" target=\"_blank\" rel=\"noreferrer noopener\">una delle minacce emergenti alla sicurezza<\/a> dei sistemi di intelligenza artificiale. Un recente lavoro accademico pubblicato su\u00a0<strong>arXiv<\/strong>\u00a0(febbraio 2026) analizza in profondit\u00e0 come le query fattuali e controfattuali possano essere sfruttate per ricostruire completamente un modello lineare \u201cblack box\u201d, mettendo a rischio propriet\u00e0 intellettuale e privacy.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Lo studio si concentra su tre tipi di interrogazioni: query fattuali, controfattuali esatte e controfattuali robuste. Gli autori dimostrano che, anche senza recuperare direttamente i parametri del modello, \u00e8 possibile delimitare matematicamente le regioni di classificazione note a partire da un insieme arbitrario di risposte. Il risultato pi\u00f9 critico riguarda per\u00f2 l\u2019estrazione completa dei parametri: con misure di distanza differenziabili (come la norma \u21132),&nbsp;<strong>una singola query controfattuale pu\u00f2 essere sufficiente<\/strong>&nbsp;per ricostruire l\u2019intero modello. Al contrario, usando distanze poliedrali (ad esempio \u21131 o \u2113\u221e), il numero di query cresce linearmente con la dimensione dei dati e raddoppia nel caso di controfattuali robusti.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Questi risultati rafforzano un punto gi\u00e0 evidenziato dalla letteratura sulla sicurezza dell\u2019AI: spiegabilit\u00e0 e protezione del modello sono spesso in tensione. Le spiegazioni controfattuali, promosse dall\u2019Explainable AI per aumentare fiducia e accountability, possono diventare vettori di attacco se non progettate con adeguate garanzie.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Il tema \u00e8 particolarmente rilevante nei settori regolamentati. Normative come il&nbsp;<strong>Comitato di Basilea per la vigilanza bancaria<\/strong>&nbsp;(BCBS), il&nbsp;<strong>Regolamento generale sulla protezione dei dati<\/strong>&nbsp;e le linee guida&nbsp;<strong>Federal Reserve System<\/strong>&nbsp;sul&nbsp;<em>model risk management<\/em>&nbsp;spingono verso modelli interpretabili, spesso lineari. Proprio questi modelli, per\u00f2, risultano pi\u00f9 facilmente estraibili.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">Approfondimento \u2013 Cronologia essenziale<\/h3>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>2015<\/strong>\u00a0\u2013 Fredrikson et al.: primi attacchi di\u00a0<em>model inversion<\/em>.<br>https:\/\/www.usenix.org\/conference\/usenixsecurity15\/technical-sessions\/presentation\/fredrikson<\/li>\n\n\n\n<li><strong>2017<\/strong>\u00a0\u2013 Wachter et al.: definizione delle spiegazioni controfattuali.<br><a href=\"https:\/\/arxiv.org\/abs\/1711.00399\">https:\/\/arxiv.org\/abs\/1711.00399<\/a><\/li>\n\n\n\n<li><strong>2019\u20132024<\/strong>\u00a0\u2013 Studi su rischi di sicurezza delle spiegazioni AI (Milli, Shokri, Nguyen).<br><a href=\"https:\/\/arxiv.org\/abs\/1907.01197\">https:\/\/arxiv.org\/abs\/1907.01197<\/a><\/li>\n\n\n\n<li><strong>2026<\/strong>\u00a0\u2013 Studio su\u00a0<em>Linear Model Extraction via Factual and Counterfactual Queries<\/em>.<br>https:\/\/arxiv.org\/abs\/2602.09748<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Consigli di approfondimento<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>NIST \u2013 Adversarial ML Threat Matrix:\u00a0<a href=\"https:\/\/www.nist.gov\/\">https:\/\/www.nist.gov<\/a><\/li>\n\n\n\n<li>ENISA \u2013 AI cybersecurity:\u00a0<a href=\"https:\/\/www.enisa.europa.eu\/\">https:\/\/www.enisa.europa.eu<\/a><\/li>\n<\/ul>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<h3 class=\"wp-block-heading\">Abstract \u2013 Pro, rischi e implicazioni sociali<\/h3>\n\n\n\n<p class=\"wp-block-paragraph\">Le spiegazioni controfattuali migliorano trasparenza e contestabilit\u00e0 delle decisioni automatiche. Tuttavia, se non mitigate, facilitano l\u2019estrazione di modelli e l\u2019esposizione di informazioni sensibili. Il rischio etico \u00e8 una normalizzazione di sistemi \u201ctrasparenti ma vulnerabili\u201d, con conseguenze su concorrenza, privacy e fiducia pubblica. In futuro, sar\u00e0 cruciale integrare spiegabilit\u00e0 e&nbsp;<em>security-by-design<\/em>, bilanciando diritto alla spiegazione e protezione dei modelli.<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"683\" src=\"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai-1024x683.jpg\" alt=\"Attacchi di model extraction: i rischi nascosti delle spiegazioni AI\" class=\"wp-image-114571\" srcset=\"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai-1024x683.jpg 1024w, https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai-300x200.jpg 300w, https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai-768x512.jpg 768w, https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai-696x464.jpg 696w, https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai-1068x712.jpg 1068w, https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg 1536w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n","protected":false},"excerpt":{"rendered":"<p>Le spiegazioni controfattuali, pensate per rendere l\u2019AI pi\u00f9 trasparente, possono diventare un\u2019arma per ricostruire modelli riservati. Un nuovo studio teorico mostra come, nel caso di modelli lineari, bastino pochissime query per comprometterne la sicurezza. L\u2019estrazione di modelli (model extraction) \u00e8 una delle minacce emergenti alla sicurezza dei sistemi di intelligenza artificiale. Un recente lavoro accademico [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":114571,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"om_disable_all_campaigns":false,"footnotes":""},"categories":[53,429,365,4,436],"tags":[215,154,63],"class_list":["post-114550","post","type-post","status-publish","format-standard","has-post-thumbnail","category-newsethos","category-ricerca-scientifica-aiopenmind","category-newscybersecurityaiopenmind","category-in-evidenza-e-7giorninews","category-unione-europea","tag-cybersecurity","tag-etica","tag-report"],"acf":[],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v28.8 - https:\/\/yoast.com\/product\/yoast-seo-wordpress\/ -->\n<title>TOPstory -\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI &#8212; AIopenmind<\/title>\n<meta name=\"description\" content=\"Attacchi di model extraction sfruttano spiegazioni AI e controfattuali: rischi per sicurezza, privacy e modelli lineari.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/\" \/>\n<meta property=\"og:locale\" content=\"it_IT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"TOPstory -\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI &#8212; AIopenmind\" \/>\n<meta property=\"og:description\" content=\"Attacchi di model extraction sfruttano spiegazioni AI e controfattuali: rischi per sicurezza, privacy e modelli lineari.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/\" \/>\n<meta property=\"og:site_name\" content=\"AIopenmind\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/aiopenmind\" \/>\n<meta property=\"article:published_time\" content=\"2026-02-13T23:58:13+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2026-05-01T12:36:07+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1536\" \/>\n\t<meta property=\"og:image:height\" content=\"1024\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"author\" content=\"adminanto\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@AIopenmind\" \/>\n<meta name=\"twitter:site\" content=\"@AIopenmind\" \/>\n<meta name=\"twitter:label1\" content=\"Scritto da\" \/>\n\t<meta name=\"twitter:data1\" content=\"adminanto\" \/>\n\t<meta name=\"twitter:label2\" content=\"Tempo di lettura stimato\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 minuti\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/\"},\"author\":{\"name\":\"adminanto\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#\\\/schema\\\/person\\\/497572e8ff03fd87ebefe4f290570051\"},\"headline\":\"TOPstory &#8211;\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI\",\"datePublished\":\"2026-02-13T23:58:13+00:00\",\"dateModified\":\"2026-05-01T12:36:07+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/\"},\"wordCount\":452,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/wp-content\\\/uploads\\\/2026\\\/02\\\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg\",\"keywords\":[\"Cybersecurity\",\"Etica\",\"Report\"],\"articleSection\":[\"#ETHOS\",\"#RICERCA SCIENTIFICA, PAPER, REPORT\",\"CYBERSECURITY\",\"IN EVIDENZA e 7GiorniNEWS\",\"UNIONE EUROPEA\"],\"inLanguage\":\"it-IT\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/\",\"url\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/\",\"name\":\"TOPstory -\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI &#8212; AIopenmind\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/wp-content\\\/uploads\\\/2026\\\/02\\\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg\",\"datePublished\":\"2026-02-13T23:58:13+00:00\",\"dateModified\":\"2026-05-01T12:36:07+00:00\",\"description\":\"Attacchi di model extraction sfruttano spiegazioni AI e controfattuali: rischi per sicurezza, privacy e modelli lineari.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#breadcrumb\"},\"inLanguage\":\"it-IT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#primaryimage\",\"url\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/wp-content\\\/uploads\\\/2026\\\/02\\\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg\",\"contentUrl\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/wp-content\\\/uploads\\\/2026\\\/02\\\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg\",\"width\":1536,\"height\":1024,\"caption\":\"Attacchi di model extraction: i rischi nascosti delle spiegazioni AI\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"TOPstory &#8211;\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#website\",\"url\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/\",\"name\":\"AIopenmind\",\"description\":\"Artificial Intelligence\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"it-IT\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#organization\",\"name\":\"AIopenmind\",\"url\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/wp-content\\\/uploads\\\/2021\\\/04\\\/aiopenmind.jpg\",\"contentUrl\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/wp-content\\\/uploads\\\/2021\\\/04\\\/aiopenmind.jpg\",\"width\":332,\"height\":217,\"caption\":\"AIopenmind\"},\"image\":{\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/aiopenmind\",\"https:\\\/\\\/x.com\\\/AIopenmind\",\"https:\\\/\\\/www.linkedin.com\\\/in\\\/aiopenmind\\\/\",\"https:\\\/\\\/www.tiktok.com\\\/@aiopenmind\",\"https:\\\/\\\/www.instagram.com\\\/aiopenmind\\\/\",\"https:\\\/\\\/www.youtube.com\\\/channel\\\/UC_0dHy6JRrolCShIZS6D0Hw\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/#\\\/schema\\\/person\\\/497572e8ff03fd87ebefe4f290570051\",\"name\":\"adminanto\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"it-IT\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/2139dfab7b92d46fd4478972312bf205fb9ddf4011d3d4d2d10d81c6221daba6?s=96&d=mm&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/2139dfab7b92d46fd4478972312bf205fb9ddf4011d3d4d2d10d81c6221daba6?s=96&d=mm&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/2139dfab7b92d46fd4478972312bf205fb9ddf4011d3d4d2d10d81c6221daba6?s=96&d=mm&r=g\",\"caption\":\"adminanto\"},\"sameAs\":[\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\"],\"url\":\"https:\\\/\\\/www.aiopenmind.it\\\/ArtificialIntelligence\\\/author\\\/adminanto\\\/\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"TOPstory -\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI &#8212; AIopenmind","description":"Attacchi di model extraction sfruttano spiegazioni AI e controfattuali: rischi per sicurezza, privacy e modelli lineari.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/","og_locale":"it_IT","og_type":"article","og_title":"TOPstory -\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI &#8212; AIopenmind","og_description":"Attacchi di model extraction sfruttano spiegazioni AI e controfattuali: rischi per sicurezza, privacy e modelli lineari.","og_url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/","og_site_name":"AIopenmind","article_publisher":"https:\/\/www.facebook.com\/aiopenmind","article_published_time":"2026-02-13T23:58:13+00:00","article_modified_time":"2026-05-01T12:36:07+00:00","og_image":[{"width":1536,"height":1024,"url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg","type":"image\/jpeg"}],"author":"adminanto","twitter_card":"summary_large_image","twitter_creator":"@AIopenmind","twitter_site":"@AIopenmind","twitter_misc":{"Scritto da":"adminanto","Tempo di lettura stimato":"3 minuti"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#article","isPartOf":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/"},"author":{"name":"adminanto","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#\/schema\/person\/497572e8ff03fd87ebefe4f290570051"},"headline":"TOPstory &#8211;\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI","datePublished":"2026-02-13T23:58:13+00:00","dateModified":"2026-05-01T12:36:07+00:00","mainEntityOfPage":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/"},"wordCount":452,"commentCount":0,"publisher":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#organization"},"image":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#primaryimage"},"thumbnailUrl":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg","keywords":["Cybersecurity","Etica","Report"],"articleSection":["#ETHOS","#RICERCA SCIENTIFICA, PAPER, REPORT","CYBERSECURITY","IN EVIDENZA e 7GiorniNEWS","UNIONE EUROPEA"],"inLanguage":"it-IT","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/","url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/","name":"TOPstory -\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI &#8212; AIopenmind","isPartOf":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#primaryimage"},"image":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#primaryimage"},"thumbnailUrl":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg","datePublished":"2026-02-13T23:58:13+00:00","dateModified":"2026-05-01T12:36:07+00:00","description":"Attacchi di model extraction sfruttano spiegazioni AI e controfattuali: rischi per sicurezza, privacy e modelli lineari.","breadcrumb":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#breadcrumb"},"inLanguage":"it-IT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/"]}]},{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#primaryimage","url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg","contentUrl":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2026\/02\/attacchi-model-extraction-rischi-spiegazioni-ai.jpg","width":1536,"height":1024,"caption":"Attacchi di model extraction: i rischi nascosti delle spiegazioni AI"},{"@type":"BreadcrumbList","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/exclusive-attacchi-di-model-extraction-i-rischi-nascosti-delle-spiegazioni-ai\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/"},{"@type":"ListItem","position":2,"name":"TOPstory &#8211;\u00a0Attacchi di model extraction: i rischi nascosti delle spiegazioni AI"}]},{"@type":"WebSite","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#website","url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/","name":"AIopenmind","description":"Artificial Intelligence","publisher":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"it-IT"},{"@type":"Organization","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#organization","name":"AIopenmind","url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/","logo":{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#\/schema\/logo\/image\/","url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2021\/04\/aiopenmind.jpg","contentUrl":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-content\/uploads\/2021\/04\/aiopenmind.jpg","width":332,"height":217,"caption":"AIopenmind"},"image":{"@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/aiopenmind","https:\/\/x.com\/AIopenmind","https:\/\/www.linkedin.com\/in\/aiopenmind\/","https:\/\/www.tiktok.com\/@aiopenmind","https:\/\/www.instagram.com\/aiopenmind\/","https:\/\/www.youtube.com\/channel\/UC_0dHy6JRrolCShIZS6D0Hw"]},{"@type":"Person","@id":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/#\/schema\/person\/497572e8ff03fd87ebefe4f290570051","name":"adminanto","image":{"@type":"ImageObject","inLanguage":"it-IT","@id":"https:\/\/secure.gravatar.com\/avatar\/2139dfab7b92d46fd4478972312bf205fb9ddf4011d3d4d2d10d81c6221daba6?s=96&d=mm&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/2139dfab7b92d46fd4478972312bf205fb9ddf4011d3d4d2d10d81c6221daba6?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/2139dfab7b92d46fd4478972312bf205fb9ddf4011d3d4d2d10d81c6221daba6?s=96&d=mm&r=g","caption":"adminanto"},"sameAs":["https:\/\/www.aiopenmind.it\/ArtificialIntelligence"],"url":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/author\/adminanto\/"}]}},"_links":{"self":[{"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/posts\/114550","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/comments?post=114550"}],"version-history":[{"count":2,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/posts\/114550\/revisions"}],"predecessor-version":[{"id":114632,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/posts\/114550\/revisions\/114632"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/media\/114571"}],"wp:attachment":[{"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/media?parent=114550"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/categories?post=114550"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.aiopenmind.it\/ArtificialIntelligence\/wp-json\/wp\/v2\/tags?post=114550"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}