{"id":14558,"date":"2026-06-02T21:06:29","date_gmt":"2026-06-02T18:06:29","guid":{"rendered":"https:\/\/www.kaspersky.com.tr\/blog\/?p=14558"},"modified":"2026-06-02T21:06:29","modified_gmt":"2026-06-02T18:06:29","slug":"llmjacking-2026-private-ai-server-security","status":"publish","type":"post","link":"https:\/\/www.kaspersky.com.tr\/blog\/llmjacking-2026-private-ai-server-security\/14558\/","title":{"rendered":"LLM ak\u0131nc\u0131lar\u0131 ve onlarla ba\u015fa \u00e7\u0131kman\u0131n yollar\u0131"},"content":{"rendered":"<p>AI g\u00fcvenli\u011fi; veri h\u0131rs\u0131zl\u0131\u011f\u0131n\u0131 \u00f6nleme, <a href=\"https:\/\/www.kaspersky.com.tr\/blog\/moltbot-enterprise-risk-management\/14307\/\" target=\"_blank\" rel=\"noopener\">hileli AI arac\u0131lar\u0131n\u0131<\/a> k\u0131s\u0131tlama veya asistanlar\u0131n zararl\u0131 tavsiyeler vermesini durdurmadan daha fazlas\u0131n\u0131 kapsar. Nispeten basit ancak h\u0131zla yayg\u0131nla\u015fan bir tehdit ortaya \u00e7\u0131kt\u0131: Hesaplama g\u00fcc\u00fcn\u00fc ele ge\u00e7irip ba\u015fkalar\u0131n\u0131n n\u00f6ral a\u011flar\u0131n\u0131 ki\u015fisel \u00e7\u0131karlar i\u00e7in kullanmaya y\u00f6nelik giri\u015fimler. Buna LLMjacking deniyor. Yapay zeka hesaplama maliyetlerinin <a href=\"https:\/\/oplexa.com\/ai-inference-cost-crisis-2026\/\" target=\"_blank\" rel=\"noopener nofollow\">b\u00fcy\u00fck \u00f6l\u00e7\u00fcde artaca\u011f\u0131<\/a> \u00f6ng\u00f6r\u00fcl\u00fcrken, bu motivasyonlarla sald\u0131r\u0131 d\u00fczenleyenlerin say\u0131s\u0131n\u0131n da artmas\u0131 bekleniyor. Sonu\u00e7 olarak, \u00f6zel yapay zeka sunucular\u0131 ve <a href=\"https:\/\/tr.wikipedia.org\/wiki\/Eri%C5%9Fim_destekli_%C3%BCretim\" target=\"_blank\" rel=\"noopener nofollow\">RAG<\/a> veya <a href=\"https:\/\/tr.wikipedia.org\/wiki\/Model_Context_Protocol\" target=\"_blank\" rel=\"noopener nofollow\">MCP<\/a> gibi bunlar\u0131 destekleyen ekosistemleri devreye al\u0131rken, ilk g\u00fcnden itibaren s\u0131k\u0131 g\u00fcvenlik \u00f6nlemleri almak hayati \u00f6nem ta\u015f\u0131yor.<\/p>\n<h2>Bir bal k\u00fcp\u00fcnden gelen istatistikler<\/h2>\n<p>Bu kaynak ele ge\u00e7irme giri\u015fimlerinin h\u0131z\u0131 ve boyutu, en iyi \u015fekilde Nisan 2026\u2019da ayr\u0131nt\u0131lar\u0131yla belgelenen bir <a href=\"https:\/\/web.archive.org\/web\/20260412230759\/https:\/www.reddit.com\/r\/ollama\/comments\/1sff7i0\/30_days_of_an_llm_honeypot\/?solution=64b4494d52a1506664b4494d52a15066&amp;js_challenge=1&amp;token=bbbe4bf1c9a2b5160829c4be34da586161e5e506c5ef56a1c8acb5184e8d115f\" target=\"_blank\" rel=\"noopener nofollow\">deneyde<\/a> ortaya kondu. Ara\u015ft\u0131rmac\u0131, bir Raspberry Pi\u2019yi y\u00fcksek performansl\u0131 \u00f6zel bir yapay zeka sunucusu gibi g\u00f6sterecek \u015fekilde yap\u0131land\u0131rd\u0131 ve internet \u00fczerinden eri\u015filebilir hale getirdi. Sorguland\u0131\u011f\u0131nda; Ollama, LM Studio, AutoGPT, LangServe ve text-gen-webui sunucular\u0131n\u0131n kullan\u0131labilir oldu\u011funu bildirdi. Bu ara\u00e7lar\u0131n t\u00fcm\u00fc, yerel olarak bar\u0131nd\u0131r\u0131lan yapay zeka modelleri i\u00e7in sarmalay\u0131c\u0131 olarak yayg\u0131n olarak kullan\u0131lan ara\u00e7lard\u0131r. Sunucu, sekt\u00f6r standard\u0131 haline gelen OpenAI format\u0131ndaki API taleplerini de kabul etmeye haz\u0131r g\u00f6r\u00fcn\u00fcyordu.<\/p>\n<p>G\u00f6r\u00fcn\u00fc\u015fe g\u00f6re t\u00fcm bu hizmetler, g\u00fcvenlik uyumu kald\u0131r\u0131lm\u0131\u015f, en g\u00fc\u00e7l\u00fc a\u00e7\u0131k kaynakl\u0131 modellerden biri olan Qwen3-Coder 30B Heretic modelinin yerel bir \u00f6rne\u011fi taraf\u0131ndan destekleniyordu. \u00dcst\u00fcne bir de bu tuzak \u00e7e\u015fitli RAG veri tabanlar\u0131n\u0131n ve <em>get_credentials <\/em>gibi cazip \u00f6zelliklere sahip bir MCP sunucusunun varl\u0131\u011f\u0131n\u0131 ortaya \u00e7\u0131kard\u0131.<\/p>\n<p>Asl\u0131nda Raspberry Pi, ger\u00e7ek bir Qwen3 modelinden \u00f6nceden kaydedilmi\u015f 500 yan\u0131t\u0131 bar\u0131nd\u0131r\u0131yordu ve gelen her sorgu i\u00e7in en alakal\u0131 cevab\u0131 se\u00e7en hafif bir komut dosyas\u0131 kullan\u0131l\u0131yordu. Bu d\u00fczenleme, y\u00fczeysel bir kontrol\u00fc ge\u00e7mek i\u00e7in yeterli olurken, ara\u015ft\u0131rmac\u0131n\u0131n sald\u0131rganlar\u0131n niyetlerini ara\u015ft\u0131rmas\u0131na da olanak sa\u011flad\u0131.<\/p>\n<p>Yazara g\u00f6re, pop\u00fcler bir internet tarama hizmeti olan Shodan, sunucunun yay\u0131na girmesinden \u00fc\u00e7 saat sonra onu tespit etti. Sadece bir saat sonra, ke\u015fif ama\u00e7l\u0131 talepler ak\u0131n ak\u0131n gelmeye ba\u015flad\u0131. Takip eden ay boyunca, sunucu binlerce farkl\u0131 IP adresinden gelen 113.000\u2019den fazla talebi i\u015fledi; bu trafi\u011fin %23\u2019\u00fc \u00f6zellikle yapay zeka yeteneklerini ke\u015ffetmeye ve yerel b\u00fcy\u00fck dil modelleri ile yapay zeka ajanlar\u0131n\u0131 istismar etmeye y\u00f6nelikti.<\/p>\n<p><em>\/api\/tags<\/em> ve <em>\/v1\/models<\/em> gibi u\u00e7 noktalara yap\u0131lan talepler, sald\u0131rganlar\u0131n sunucuda hangi modellerin bar\u0131nd\u0131r\u0131ld\u0131\u011f\u0131n\u0131 tespit etmelerine olanak tan\u0131rken, <em>\/.cursor\/rules<\/em> adresini taramak genellikle bir yapay zeka ajan\u0131 \u00fczerinde istismar giri\u015fiminden \u00f6nce gelir. Benzer \u015fekilde, <em>\/.well-known\/mcp.json<\/em> dosyas\u0131n\u0131 kontrol etmek, kurban\u0131n MCP sunucular\u0131n\u0131n bir listesini elde etmeyi sa\u011flar. Yazar, basit tarama a\u015famas\u0131n\u0131 a\u015fan sald\u0131r\u0131lar\u0131n toplam say\u0131s\u0131na de\u011finmese de, yaln\u0131zca deneyin son haftas\u0131nda LLM\u2019yi ele ge\u00e7irmeye y\u00f6nelik 175 aktif giri\u015fim ger\u00e7ekle\u015fti.<\/p>\n<h2>Sald\u0131rganlar\u0131n amac\u0131 ne?<\/h2>\n<p>Ara\u015ft\u0131rmac\u0131n\u0131n g\u00f6zlemlerine g\u00f6re, yem sunucusunu hedef alan sald\u0131r\u0131lar\u0131n hi\u00e7biri rastgele kod \u00e7al\u0131\u015ft\u0131rmaya veya k\u00f6k eri\u015fimi elde etmeye \u00e7al\u0131\u015fmad\u0131. (Edit\u00f6r notu: Bu durum \u015fa\u015f\u0131rt\u0131c\u0131d\u0131r ve g\u00fcnl\u00fck kayd\u0131 sistemindeki eksikliklere i\u015faret ediyor olabilir.) Sald\u0131r\u0131lar\u0131n neredeyse tamam\u0131 kaynaklar\u0131 t\u00fcketmeyi ama\u00e7l\u0131yordu. \u00d6rne\u011fin, deney s\u0131ras\u0131nda \u015fu etkinlikler kaydedildi:<\/p>\n<ul>\n<li>Bir mikroi\u015flemcinin teknik belgelerini incelemek i\u00e7in iyi yap\u0131land\u0131r\u0131lm\u0131\u015f bir giri\u015fim<\/li>\n<li>Erotik bir roman yazma istemi<\/li>\n<li>Yeni g\u00fcvenlik a\u00e7\u0131klar\u0131yla ilgili sosyal medya metin verilerinin ayr\u0131\u015ft\u0131r\u0131lmas\u0131 ve yap\u0131land\u0131r\u0131lmas\u0131na y\u00f6nelik talepler<\/li>\n<li>G\u00fcvenli\u011fi ihlal edilmi\u015f sunucuyu bir API proxy\u2019si olarak kullanarak Anthropic modellerini \u00e7a\u011f\u0131rma giri\u015fimi<\/li>\n<\/ul>\n<p>Yapay zeka kaynaklar\u0131n\u0131n ke\u015ffinde standartla\u015ft\u0131r\u0131lm\u0131\u015f ve h\u0131zla geli\u015fen ara\u00e7lar\u0131n kullan\u0131ld\u0131\u011f\u0131na dikkat \u00e7ekmek gerekir. LLM-Scanner adl\u0131 bir uygulamadan gelen talepler, sekiz \u00fclkedeki yedi farkl\u0131 bulut sa\u011flay\u0131c\u0131s\u0131n\u0131n altyap\u0131s\u0131ndan kaynaklan\u0131yordu; bu durum, sald\u0131rganlar\u0131n yerle\u015fik y\u00f6ntemlerin yan\u0131 s\u0131ra teknikleri payla\u015fmaya y\u00f6nelik \u00f6zel platformlar da kurduklar\u0131n\u0131 g\u00f6steriyor. Deneyin \u00fc\u00e7\u00fcnc\u00fc haftas\u0131na gelindi\u011finde, taray\u0131c\u0131ya ek bir kontrol mekanizmas\u0131 eklenmi\u015fti: Art\u0131k, kar\u015f\u0131s\u0131ndaki ki\u015finin canl\u0131 bir yapay zeka m\u0131 yoksa \u00f6nceden haz\u0131rlanm\u0131\u015f yan\u0131tlar veren bir tuzak m\u0131 oldu\u011funu belirlemek i\u00e7in basit ve soyut sorular kullan\u0131yordu.<\/p>\n<p>Spesifik olmayan sald\u0131r\u0131lar aras\u0131nda, deneyde <em>.env<\/em> dosyas\u0131ndan kimlik bilgilerini ele ge\u00e7irmeye y\u00f6nelik \u00e7ok say\u0131da giri\u015fim kaydedildi. Sald\u0131rganlar, sunucudaki akla gelebilecek her dizinde bu dosyay\u0131 sistematik olarak arad\u0131lar. <em>.env<\/em> dosyas\u0131n\u0131 herkesin eri\u015fimine a\u00e7\u0131k b\u0131rakmak, Laravel, <em>Node.js<\/em> ve di\u011fer \u00e7er\u00e7evelerde proje da\u011f\u0131t\u0131m\u0131 yaparken yap\u0131lan en temel hatalardan biridir; ancak bu, \u00f6zellikle yeni ba\u015flayanlar ve acemi geli\u015ftiriciler aras\u0131nda h\u00e2l\u00e2 s\u0131k\u00e7a g\u00f6zden ka\u00e7an bir durumdur. Sonu\u00e7 olarak, sald\u0131rganlar\u0131n \u00e7abalar\u0131n\u0131n kar\u015f\u0131l\u0131\u011f\u0131n\u0131 alacaklar\u0131n\u0131 ummak i\u00e7in her t\u00fcrl\u00fc nedenleri var.<\/p>\n<h2>Sonu\u00e7lar ve savunma ipu\u00e7lar\u0131<\/h2>\n<p>Halka a\u00e7\u0131k sunucular\u0131 taramak ve bunlar\u0131 istismar etmeye \u00e7al\u0131\u015fmak yeni bir \u015fey de\u011fil, ancak b\u00fcy\u00fck dil modellerinin (LLM) yayg\u0131nla\u015fmas\u0131, sald\u0131rganlara \u00e7abalar\u0131ndan para kazanman\u0131n yeni bir yolunu sunuyor. Bu yol, sald\u0131rganlar i\u00e7in son derece kazan\u00e7l\u0131, kurbanlar i\u00e7inse y\u0131k\u0131c\u0131 sonu\u00e7lar do\u011furuyor. Bu sald\u0131r\u0131lar\u0131n ne kadar b\u00fcy\u00fck boyutlara ula\u015fabilece\u011fini anlamak i\u00e7in, en yak\u0131n benzeri olan kripto madencili\u011fi h\u0131rs\u0131zl\u0131\u011f\u0131 pazar\u0131na bak\u0131n. Burada su\u00e7lular, \u00e7al\u0131nan bilgi i\u015flem kaynaklar\u0131n\u0131 kullanarak kripto para madencili\u011fi yapar. S\u00f6z konusu pazar, <a href=\"https:\/\/www.economist.com\/science-and-technology\/2026\/04\/22\/crypto-miners-are-quietly-colonising-computers\" target=\"_blank\" rel=\"noopener nofollow\">yaln\u0131zca 2025 y\u0131l\u0131nda %20 b\u00fcy\u00fcd\u00fc<\/a>. Yapay zeka destekli \u00e7\u00f6z\u00fcmlerin yayg\u0131nla\u015fmas\u0131, b\u00fcy\u00fck sa\u011flay\u0131c\u0131lar\u0131n abonelik \u00fccretlerini art\u0131rmas\u0131 ve yerel yapay zeka yongalar\u0131n\u0131n arz\u0131n\u0131n yetersiz kalmas\u0131yla birlikte, LLMjacking\u2019in end\u00fcstriyel \u00f6l\u00e7ekte bir fenomen haline gelmesini beklemeliyiz.<\/p>\n<p>\u00d6zel yapay zeka altyap\u0131s\u0131 i\u00e7in temel g\u00fcvenlik \u00f6nlemleri<\/p>\n<ul>\n<li>Tek bir makinede yerel olarak \u00e7al\u0131\u015fan yapay zeka sistemleri i\u00e7in, LM Studio, Ollama veya benzeri sunucular\u0131n, mevcut t\u00fcm a\u011f arabirimleri yerine yaln\u0131zca yerel arabirimden (localhost) ba\u011flant\u0131 kabul edecek \u015fekilde yap\u0131land\u0131r\u0131ld\u0131\u011f\u0131ndan emin olun. Bu, LLM\u2019nin eri\u015fimini ana makineyle s\u0131n\u0131rlar ve yapay zekan\u0131n internet \u00fczerinden eri\u015filebilir olmas\u0131n\u0131 engeller.<\/li>\n<li>Uzak talepleri i\u015fleyen sunucular i\u00e7in (sunucu yaln\u0131zca yerel bir kurumsal a\u011f i\u00e7inde \u00e7al\u0131\u015f\u0131yor olsa bile), yaln\u0131zca API anahtar\u0131 do\u011frulamas\u0131na g\u00fcvenmek yerine <a href=\"https:\/\/www.kaspersky.com.tr\/blog\/how-to-benefit-from-identity-security\/11646\/\" target=\"_blank\" rel=\"noopener\">sa\u011flam kimlik do\u011frulama ve yetkilendirme mekanizmalar\u0131<\/a> uygulay\u0131n. K\u0131sa \u00f6m\u00fcrl\u00fc belirte\u00e7lere dayal\u0131 OIDC veya OAuth2 tabanl\u0131 \u00e7\u00f6z\u00fcmler en etkili olanlard\u0131r. Bu, sadece LLMjacking\u2019e kar\u015f\u0131 koruma sa\u011flamakla kalmaz, ayn\u0131 zamanda kullan\u0131c\u0131 faaliyetlerinin daha ayr\u0131nt\u0131l\u0131 bir \u015fekilde izlenmesini sa\u011flar ve API anahtarlar\u0131n\u0131n k\u00f6t\u00fcye kullan\u0131lmas\u0131n\u0131 \u00f6nler. Ayr\u0131ca, anahtarlar yaln\u0131zca d\u0131\u015f sald\u0131r\u0131lara kar\u015f\u0131 koruma sa\u011flamakla kalmamal\u0131; <a href=\"https:\/\/www.theregister.com\/2026\/04\/27\/cursoropus_agent_snuffs_out_pocketos\/\" target=\"_blank\" rel=\"noopener nofollow\">yapay zeka ajanlar\u0131n\u0131n kendileri taraf\u0131ndan anahtarlar\u0131n k\u00f6t\u00fcye kullan\u0131lmas\u0131 <\/a>da giderek artan bir risk olu\u015fturmaktad\u0131r. Bu durum, LLM arabirimlerinin yan\u0131 s\u0131ra MCP, RAG ve di\u011ferleri i\u00e7in de ge\u00e7erlidir.<\/li>\n<li>A\u011f segmentasyonu ve IP izin listelerini kullanarak, yapay zeka sunucusuna yaln\u0131zca buna ihtiya\u00e7 duyan departmanlara, \u00e7al\u0131\u015fanlara ve hizmetlere eri\u015fim izni verin.<\/li>\n<li>T\u00fcm istemci-sunucu ba\u011flant\u0131lar\u0131n\u0131n g\u00fcncel bir TLS s\u00fcr\u00fcm\u00fcyle g\u00fcvenli hale getirildi\u011finden emin olun.<\/li>\n<li>Belirli hizmetlere eri\u015fimi birbirinden ay\u0131rarak <a href=\"https:\/\/www.kaspersky.com\/blog\/what-is-the-principle-of-least-privilege\/50232\/\" target=\"_blank\" rel=\"noopener nofollow\">en az ayr\u0131cal\u0131k ilkesini <\/a>uygulay\u0131n; \u00f6rne\u011fin, MCP ve LLM bile\u015fenlerinin kendilerine \u00f6zg\u00fc belirte\u00e7leri olmal\u0131d\u0131r.<\/li>\n<li>Yapay zeka modellerini bar\u0131nd\u0131ranlar da dahil olmak \u00fczere t\u00fcm i\u015f istasyonlar\u0131na ve sunuculara bir <a href=\"https:\/\/www.kaspersky.com.tr\/enterprise-security\/endpoint-detection-response-edr?icid=tr_kdailyplacehold_acq_ona_smm__onl_b2b_kasperskydaily_wpplaceholder_______\" target=\"_blank\" rel=\"noopener\">EDR Agent<\/a> y\u00fcklendi\u011finden emin olun.<\/li>\n<li>Yapay zeka kaynak kullan\u0131m\u0131n\u0131 izleyin, farkl\u0131 \u00e7al\u0131\u015fan rolleri i\u00e7in kullan\u0131m kotalar\u0131 belirleyin ve ola\u011fand\u0131\u015f\u0131 etkinlik art\u0131\u015flar\u0131 i\u00e7in uyar\u0131lar ayarlay\u0131n.<\/li>\n<li>Modele ve destekleyici ara\u00e7lara y\u00f6neltilen LLM yan\u0131tlar\u0131 ile taleplerin ayr\u0131nt\u0131l\u0131 kay\u0131tlar\u0131n\u0131 tutun. Bu veri kaynaklar\u0131n\u0131 <a href=\"https:\/\/www.kaspersky.com\/enterprise-security\/managed-detection-and-response?icid=tr_kdailyplacehold_acq_ona_smm__onl_b2b_kasperskydaily_wpplaceholder_______\" target=\"_blank\" rel=\"noopener nofollow\">SIEM sisteminizle<\/a> entegre edin. G\u00fcnl\u00fcklerin tahrif edilmeye veya silinmeye kar\u015f\u0131 dayan\u0131kl\u0131 olmas\u0131n\u0131 sa\u011flay\u0131n.<\/li>\n<\/ul>\n<input type=\"hidden\" class=\"category_for_banner\" value=\"mdr\"><input type=\"hidden\" class=\"placeholder_for_banner\" data-cat_id=\"mdr\" value=\"13478\">\n","protected":false},"excerpt":{"rendered":"<p>Yapay zeka kaynaklar\u0131n\u0131 ele ge\u00e7irme giri\u015fimleri art\u0131k end\u00fcstriyel \u00f6l\u00e7ekte ger\u00e7ekle\u015fiyor. Yapay zeka altyap\u0131s\u0131 nas\u0131l hedef al\u0131n\u0131yor ve hangi savunma \u00f6nlemlerini almal\u0131s\u0131n\u0131z?<\/p>\n","protected":false},"author":2722,"featured_media":14559,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1351],"tags":[2862,744,1610,2142,1424],"class_list":{"0":"post-14558","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-threats","8":"tag-buyuk-dil-modelleri","9":"tag-guvenlik","10":"tag-makine-ogrenimi","11":"tag-sunucular","12":"tag-yapay-zeka"},"hreflang":[{"hreflang":"tr","url":"https:\/\/www.kaspersky.com.tr\/blog\/llmjacking-2026-private-ai-server-security\/14558\/"},{"hreflang":"en-in","url":"https:\/\/www.kaspersky.co.in\/blog\/llmjacking-2026-private-ai-server-security\/30714\/"},{"hreflang":"en-ae","url":"https:\/\/me-en.kaspersky.com\/blog\/llmjacking-2026-private-ai-server-security\/25767\/"},{"hreflang":"en-gb","url":"https:\/\/www.kaspersky.co.uk\/blog\/llmjacking-2026-private-ai-server-security\/30564\/"},{"hreflang":"es","url":"https:\/\/www.kaspersky.es\/blog\/llmjacking-2026-private-ai-server-security\/32133\/"},{"hreflang":"it","url":"https:\/\/www.kaspersky.it\/blog\/llmjacking-2026-private-ai-server-security\/30696\/"},{"hreflang":"ru","url":"https:\/\/www.kaspersky.ru\/blog\/llmjacking-2026-private-ai-server-security\/41840\/"},{"hreflang":"x-default","url":"https:\/\/www.kaspersky.com\/blog\/llmjacking-2026-private-ai-server-security\/55768\/"},{"hreflang":"pt-br","url":"https:\/\/www.kaspersky.com.br\/blog\/llmjacking-2026-private-ai-server-security\/25008\/"},{"hreflang":"de","url":"https:\/\/www.kaspersky.de\/blog\/llmjacking-2026-private-ai-server-security\/33508\/"},{"hreflang":"ru-kz","url":"https:\/\/blog.kaspersky.kz\/llmjacking-2026-private-ai-server-security\/30650\/"},{"hreflang":"en-au","url":"https:\/\/www.kaspersky.com.au\/blog\/llmjacking-2026-private-ai-server-security\/36224\/"},{"hreflang":"en-za","url":"https:\/\/www.kaspersky.co.za\/blog\/llmjacking-2026-private-ai-server-security\/36116\/"}],"acf":[],"banners":"","maintag":{"url":"https:\/\/www.kaspersky.com.tr\/blog\/tag\/yapay-zeka\/","name":"yapay zeka"},"_links":{"self":[{"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/posts\/14558","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/users\/2722"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/comments?post=14558"}],"version-history":[{"count":3,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/posts\/14558\/revisions"}],"predecessor-version":[{"id":14561,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/posts\/14558\/revisions\/14561"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/media\/14559"}],"wp:attachment":[{"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/media?parent=14558"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/categories?post=14558"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kaspersky.com.tr\/blog\/wp-json\/wp\/v2\/tags?post=14558"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}