{"id":990430,"date":"2025-07-29T07:00:00","date_gmt":"2025-07-29T04:00:00","guid":{"rendered":"https:\/\/gpss.ro\/security-news\/cum-se-apara-microsoft-impotriva-atacurilor-de-injectie-prompta-indirecta-3\/"},"modified":"2025-10-12T12:59:13","modified_gmt":"2025-10-12T09:59:13","slug":"cum-se-apara-microsoft-impotriva-atacurilor-de-injectie-prompta-indirecta-3","status":"publish","type":"security_news","link":"https:\/\/delve.ro\/ro\/security-news\/cum-se-apara-microsoft-impotriva-atacurilor-de-injectie-prompta-indirecta-3\/","title":{"rendered":"How Microsoft defends against indirect prompt injection attacks"},"content":{"rendered":"<div class=\"gpss-language-switcher\" style=\"margin-bottom: 20px; padding: 15px; background: #f0f9ff; border-left: 4px solid #3b82f6; border-radius: 8px;\">\n            <div style=\"display: flex; align-items: center; justify-content: space-between; flex-wrap: wrap; gap: 10px;\">\n                <div style=\"display: flex; align-items: center; gap: 10px;\">\n                    <span style=\"font-weight: 600; color: #1e40af;\">\ud83c\udf0d Limb\u0103 \/ Language:<\/span>\n                    <button onclick=\"switchLanguage('en')\" id=\"btn-lang-en\" class=\"lang-btn lang-btn-active\" style=\"padding: 8px 16px; background: #3b82f6; color: white; border: none; border-radius: 6px; cursor: pointer; font-weight: 600; transition: all 0.3s;\">\n                        \ud83c\uddec\ud83c\udde7 English (Original)\n                    <\/button>\n                    <button onclick=\"switchLanguage('ro')\" id=\"btn-lang-ro\" class=\"lang-btn\" style=\"padding: 8px 16px; background: #e5e7eb; color: #374151; border: none; border-radius: 6px; cursor: pointer; font-weight: 600; transition: all 0.3s;\">\n                        \ud83c\uddf7\ud83c\uddf4 Rom\u00e2n\u0103\n                    <\/button>\n                <\/div>\n                <small style=\"color: #6b7280; font-style: italic;\">Traducere automat\u0103 \/ Automatic translation<\/small>\n            <\/div>\n        <\/div>\n\n        <div id=\"content-en\" class=\"lang-content\" style=\"display: block;\">\n            <div class=\"article-content\">Summary The growing adoption of large language models (LLMs) in enterprise workflows has introduced a new class of adversarial techniques: indirect prompt injection. Indirect prompt injection can be used against systems that leverage large language models (LLMs) to process untrusted data. Fundamentally, the risk is that an attacker could provide specially crafted data that the LLM misinterprets as instructions.<\/div>\n        <\/div>\n\n        <div id=\"content-ro\" class=\"lang-content\" style=\"display: none;\">\n            <div class=\"article-content\"><?xml encoding=\"UTF-8\"><p>Rezumat Adoptarea din ce &icirc;n ce mai mare a modelelor de limbaj mare (LLM) &icirc;n fluxurile de lucru pentru &icirc;ntreprinderi a introdus o nou&#259; clas&#259; de tehnici adverse: injec&#539;ie prompt&#259; indirect&#259;. Injec&#539;ia prompt&#259; indirect&#259; poate fi utilizat&#259; &icirc;mpotriva sistemelor care utilizeaz&#259; modele de limbaj mare (LLMS) pentru a prelucra date de &icirc;ncredere. &Icirc;n mod fundamental, riscul este ca un atacator s&#259; poat&#259; furniza date special concepute pe care LLM le interpreteaz&#259; gre&#537;it ca instruc&#539;iuni.<\/p><\/div>\n        <\/div>\n\n        <script>\n        function switchLanguage(lang) {\n            \/\/ Ascunde ambele versiuni\n            document.getElementById(\"content-ro\").style.display = \"none\";\n            document.getElementById(\"content-en\").style.display = \"none\";\n\n            \/\/ Reseteaz\u0103 stilurile butoanelor\n            document.querySelectorAll(\".lang-btn\").forEach(function(btn) {\n                btn.style.background = \"#e5e7eb\";\n                btn.style.color = \"#374151\";\n                btn.classList.remove(\"lang-btn-active\");\n            });\n\n            \/\/ Afi\u0219eaz\u0103 versiunea selectat\u0103\n            if (lang === \"ro\") {\n                document.getElementById(\"content-ro\").style.display = \"block\";\n                document.getElementById(\"btn-lang-ro\").style.background = \"#3b82f6\";\n                document.getElementById(\"btn-lang-ro\").style.color = \"white\";\n                document.getElementById(\"btn-lang-ro\").classList.add(\"lang-btn-active\");\n            } else {\n                document.getElementById(\"content-en\").style.display = \"block\";\n                document.getElementById(\"btn-lang-en\").style.background = \"#3b82f6\";\n                document.getElementById(\"btn-lang-en\").style.color = \"white\";\n                document.getElementById(\"btn-lang-en\").classList.add(\"lang-btn-active\");\n            }\n\n            \/\/ Salveaz\u0103 preferin\u021ba \u00een localStorage\n            localStorage.setItem(\"gpss_preferred_language\", lang);\n        }\n\n        \/\/ Restaureaz\u0103 preferin\u021ba utilizatorului la \u00eenc\u0103rcare\n        document.addEventListener(\"DOMContentLoaded\", function() {\n            var preferredLang = localStorage.getItem(\"gpss_preferred_language\") || \"ro\";\n            switchLanguage(preferredLang);\n        });\n\n        \/\/ Hover effects pentru butoane\n        document.querySelectorAll(\".lang-btn\").forEach(function(btn) {\n            btn.addEventListener(\"mouseenter\", function() {\n                if (!this.classList.contains(\"lang-btn-active\")) {\n                    this.style.background = \"#bfdbfe\";\n                    this.style.color = \"#1e40af\";\n                }\n            });\n            btn.addEventListener(\"mouseleave\", function() {\n                if (!this.classList.contains(\"lang-btn-active\")) {\n                    this.style.background = \"#e5e7eb\";\n                    this.style.color = \"#374151\";\n                }\n            });\n        });\n        <\/script>\n\n        <style>\n        .lang-btn:hover {\n            transform: translateY(-2px);\n            box-shadow: 0 4px 12px rgba(59, 130, 246, 0.3);\n        }\n        .lang-btn-active {\n            box-shadow: 0 4px 12px rgba(59, 130, 246, 0.4);\n        }\n        <\/style>","protected":false},"excerpt":{"rendered":"<p>\ud83c\udf0d Limb\u0103 \/ Language: \ud83c\uddec\ud83c\udde7 English (Original) \ud83c\uddf7\ud83c\uddf4 Rom\u00e2n\u0103 Traducere automat\u0103 \/ Automatic translation Summary The growing adoption of large language models (LLMs) in enterprise workflows has introduced a new class of adversarial techniques: indirect prompt injection. Indirect prompt injection can be used against systems that leverage large language models (LLMs) to process untrusted data. [&hellip;]<\/p>\n","protected":false},"featured_media":0,"template":"","meta":{"footnotes":""},"news_region":[152],"news_source":[165],"news_severity":[146],"class_list":["post-990430","security_news","type-security_news","status-publish","hentry","news_region-international","news_source-microsoft","news_severity-low"],"_links":{"self":[{"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/security-news-items\/990430","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/security-news-items"}],"about":[{"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/types\/security_news"}],"version-history":[{"count":1,"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/security-news-items\/990430\/revisions"}],"predecessor-version":[{"id":990887,"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/security-news-items\/990430\/revisions\/990887"}],"wp:attachment":[{"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/media?parent=990430"}],"wp:term":[{"taxonomy":"news_region","embeddable":true,"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/news_region?post=990430"},{"taxonomy":"news_source","embeddable":true,"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/news_source?post=990430"},{"taxonomy":"news_severity","embeddable":true,"href":"https:\/\/delve.ro\/ro\/wp-json\/wp\/v2\/news_severity?post=990430"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}