Finans bakanlar\u0131, merkez bankac\u0131lar\u0131 ve finans\u00f6rler, finansal sistemlerin g\u00fcvenli\u011fini zay\u0131flatabilece\u011finden korktuklar\u0131 g\u00fc\u00e7l\u00fc yeni bir yapay zeka modeli hakk\u0131nda ciddi endi\u015felerini dile getirdiler.<\/p>\n
Anthropic<\/b> taraf\u0131ndan geli\u015ftirilen Claude Mythos<\/b> modelinin, bir\u00e7ok b\u00fcy\u00fck i\u015fletim sistemindeki g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 bulmas\u0131n\u0131n ard\u0131ndan kriz toplant\u0131lar\u0131na yol a\u00e7t\u0131\u011f\u0131 belirtildi.<\/p>\n
Uzmanlar, modelin siber g\u00fcvenlik zay\u0131fl\u0131klar\u0131n\u0131 tespit etme ve kullanma konusunda e\u015fi benzeri g\u00f6r\u00fclmemi\u015f bir yetene\u011fe sahip olabilece\u011fini s\u00f6ylerken, di\u011ferleri yeteneklerini tam olarak anlamak i\u00e7in daha fazla test yap\u0131lmas\u0131 gerekti\u011fini belirtiyor.<\/p>\n
Kanada Maliye Bakan\u0131 Fran\u00e7ois-Philippe Champagne<\/b>, BBC’ye verdi\u011fi deme\u00e7te, Mythos<\/b>‘un bu hafta Washington DC’deki Uluslararas\u0131 Para Fonu (IMF)<\/b> toplant\u0131s\u0131nda kapsaml\u0131 bir \u015fekilde tart\u0131\u015f\u0131ld\u0131\u011f\u0131n\u0131 s\u00f6yledi.<\/p>\n
“Kesinlikle t\u00fcm finans bakanlar\u0131n\u0131n dikkatini \u00e7ekecek kadar ciddi,” dedi.<\/p>\n
“Fark \u015fu ki, H\u00fcrm\u00fcz Bo\u011faz\u0131’n\u0131n nerede oldu\u011funu ve ne kadar b\u00fcy\u00fck oldu\u011funu biliyoruz… Anthropic<\/b> ile kar\u015f\u0131 kar\u015f\u0131ya oldu\u011fumuz sorun, bilinmeyenin, bilinmeyen olmas\u0131d\u0131r.” diye ekledi.<\/p>\n
“Bu, finansal sistemlerimizin dayan\u0131kl\u0131l\u0131\u011f\u0131n\u0131 sa\u011flamak i\u00e7in g\u00fcvencelere ve s\u00fcre\u00e7lere sahip olmam\u0131z i\u00e7in \u00e7ok fazla dikkat gerektiriyor,” diye ekledi.<\/p>\n
Mythos<\/b>, Anthropic<\/b>‘in OpenAI<\/b>‘nin ChatGPT<\/b>‘si ve Google<\/b>‘\u0131n Gemini<\/b>‘sine rakip olan daha geni\u015f yapay zeka sistemi Claude<\/b>‘un bir par\u00e7as\u0131 olarak geli\u015ftirilen en son modellerinden biridir.<\/p>\n
Bu ay\u0131n ba\u015flar\u0131nda Anthropic<\/b> taraf\u0131ndan ortaya \u00e7\u0131kar\u0131lan model, yapay zeka modellerini ve “yanl\u0131\u015f hizalanm\u0131\u015f” g\u00f6revlerdeki performanslar\u0131n\u0131 test etmekten sorumlu geli\u015ftiriciler taraf\u0131ndan “bilgisayar g\u00fcvenli\u011fi g\u00f6revlerinde \u015fa\u015f\u0131rt\u0131c\u0131 derecede yetenekli” olarak tan\u0131mland\u0131.<\/p>\n
Eski yaz\u0131l\u0131m hatalar\u0131n\u0131 ortaya \u00e7\u0131karabilece\u011fi veya sistem g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 kolayca kullanman\u0131n yollar\u0131n\u0131 bulabilece\u011fi endi\u015felerini gerek\u00e7e g\u00f6stererek, Anthropic<\/b> modeli yay\u0131nlamad\u0131.<\/p>\n
Bunun yerine, Mythos<\/b>‘u Amazon Web Services<\/b>, CrowdStrike<\/b>, Microsoft<\/b> ve Nvidia<\/b> gibi teknoloji devlerine, “d\u00fcnyan\u0131n en kritik yaz\u0131l\u0131mlar\u0131n\u0131 g\u00fcvence alt\u0131na alma \u00e7abas\u0131” olarak adland\u0131rd\u0131\u011f\u0131 Project Glasswing<\/b> adl\u0131 bir giri\u015fimin par\u00e7as\u0131 olarak sundu.<\/p>\n
Per\u015fembe g\u00fcn\u00fc, Anthropic<\/b> mevcut bir model olan Claude Opus<\/b>‘un yeni bir s\u00fcr\u00fcm\u00fcn\u00fc yay\u0131nlayarak, Mythos<\/b>‘un siber yeteneklerinin daha az g\u00fc\u00e7l\u00fc sistemlerde test edilmesine olanak tan\u0131yaca\u011f\u0131n\u0131 belirtti.<\/p>\n
Mythos<\/b> hakk\u0131nda dile getirilen endi\u015feler, \u00f6nceki yapay zeka modelleri hakk\u0131ndaki tart\u0131\u015fmalar\u0131 a\u015fabilir, ancak baz\u0131 siber g\u00fcvenlik uzmanlar\u0131, modelin ger\u00e7ekte ne kadar yetenekli oldu\u011funu g\u00f6rmek i\u00e7in daha geni\u015f end\u00fcstri taraf\u0131ndan test edilmedi\u011fi g\u00f6z \u00f6n\u00fcne al\u0131nd\u0131\u011f\u0131nda, bu endi\u015felerin ne kadar hakl\u0131 oldu\u011funu sorgulad\u0131.<\/p>\n
Birle\u015fik Krall\u0131k Yapay Zeka G\u00fcvenlik Enstit\u00fcs\u00fc<\/b>‘ne modelin \u00f6nizleme s\u00fcr\u00fcm\u00fcne eri\u015fim verildi ve modelin siber g\u00fcvenlik becerileri hakk\u0131nda tek ba\u011f\u0131ms\u0131z raporu yay\u0131nlad\u0131.<\/p>\n
Ara\u015ft\u0131rmac\u0131lar, modelin savunmas\u0131z ortamlarda bir\u00e7ok g\u00fcvenlik a\u00e7\u0131\u011f\u0131 bulabilen g\u00fc\u00e7l\u00fc bir ara\u00e7 oldu\u011funu belirtirken, Mythos<\/b>‘un Claude<\/b>‘un selefi Opus 4<\/b>‘ten dramatik bir \u015fekilde daha iyi olmad\u0131\u011f\u0131n\u0131 \u00f6ne s\u00fcrd\u00fcler.<\/p>\n
“Testlerimiz, Mythos Preview<\/b>‘\u0131n zay\u0131f g\u00fcvenlik duru\u015funa sahip sistemleri kullanabilece\u011fini ve bu yeteneklere sahip daha fazla modelin geli\u015ftirilmesinin muhtemel oldu\u011funu g\u00f6steriyor,” dedi rapor yazarlar\u0131.<\/p>\n
Bu ayn\u0131 zamanda bir yapay zeka geli\u015ftiricisinin modellerinin yeteneklerinin yay\u0131nlanmamas\u0131 gerekti\u011fi iddias\u0131nda bulundu\u011fu ilk sefer de\u011fil \u2013 ele\u015ftirmenlerin bir tan\u0131t\u0131m takti\u011fi oldu\u011funu savundu\u011fu bir durum.<\/p>\n
\u015eubat 2019’da OpenAI<\/b>, \u015fu anda en b\u00fcy\u00fck arac\u0131 ChatGPT<\/b>‘ye g\u00fc\u00e7 veren modellerinin daha \u00f6nceki bir s\u00fcr\u00fcm\u00fc olan GPT-2<\/b>‘nin yay\u0131n\u0131n\u0131 ertelemeyi se\u00e7ti\u011finde benzer korkular\u0131 dile getirmi\u015fti.<\/p>\nG\u00fcvenlik A\u00e7\u0131klar\u0131n\u0131 Anlamak<\/h3>\n
\u00dcst d\u00fczey bankac\u0131lara, sistemlerini test etmeleri i\u00e7in modele \u00f6nceden eri\u015fim sa\u011flanacak.<\/p>\n
Barclays<\/b> CEO’su CS Venkatakrishnan<\/b>, BBC’ye \u015funlar\u0131 s\u00f6yledi: “\u0130nsanlar\u0131n endi\u015felenmesi i\u00e7in yeterince ciddi.”<\/p>\n
“Onu daha iyi anlamal\u0131 ve ortaya \u00e7\u0131kan g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 anlamal\u0131 ve h\u0131zl\u0131 bir \u015fekilde d\u00fczeltmeliyiz.” diye ekledi.<\/p>\n
“Yeni d\u00fcnya bu olacak” diyerek, hem f\u0131rsatlar\u0131 hem de g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 i\u00e7eren \u00e7ok daha ba\u011flant\u0131l\u0131 bir finansal sisteme at\u0131fta bulundu.<\/p>\n
Geli\u015ftirici Anthropic<\/b>, modelin baz\u0131 kritik i\u015fletim sistemlerinde, finansal sistemlerde ve web taray\u0131c\u0131lar\u0131nda zaten birden fazla g\u00fcvenlik a\u00e7\u0131\u011f\u0131n\u0131 ortaya \u00e7\u0131kard\u0131\u011f\u0131n\u0131 belirtirken, h\u00fck\u00fcmetlere ve bankalara kendi sistemlerini korumalar\u0131na yard\u0131mc\u0131 olmak i\u00e7in halka a\u00e7\u0131k s\u00fcr\u00fcm\u00fcnden \u00f6nce eri\u015fim sunuluyor.<\/p>\n
\u0130ngiltere Merkez Bankas\u0131 Ba\u015fkan\u0131 Andrew Bailey<\/b>, BBC’ye bu geli\u015fmenin \u00e7ok ciddiye al\u0131nmas\u0131 gerekti\u011fini s\u00f6yledi: “Bu en son yapay zeka geli\u015fiminin siber su\u00e7 riski i\u00e7in ne anlama gelebilece\u011fini \u015fimdi \u00e7ok dikkatli bir \u015fekilde incelememiz gerekiyor.”<\/p>\n
“Sonu\u00e7, \u00e7ekirdek BT sistemlerindeki mevcut g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 tespit etmeyi kolayla\u015ft\u0131ran bir yapay zeka, modelleme geli\u015fimi olabilir ve ard\u0131ndan siber su\u00e7lular \u2013 k\u00f6t\u00fc niyetli akt\u00f6rler \u2013 bunlar\u0131 kullanmaya \u00e7al\u0131\u015fabilir,” diye ekledi.<\/p>\n
ABD Hazine Bakanl\u0131\u011f\u0131<\/b>, Anthropic<\/b> taraf\u0131ndan Mythos<\/b>‘un herhangi bir halka a\u00e7\u0131k s\u00fcr\u00fcm\u00fcnden \u00f6nce b\u00fcy\u00fck bankalar\u0131n\u0131 sistemlerini test etmeye te\u015fvik ederek konuyu g\u00fcndeme getirdi\u011fini do\u011frulad\u0131.<\/p>\n
Finans sekt\u00f6r\u00fc kaynaklar\u0131, ba\u015fka bir \u00f6nde gelen ABD yapay zeka \u015firketinin yak\u0131nda benzer derecede g\u00fc\u00e7l\u00fc ancak ayn\u0131 g\u00fcvencelere sahip olmayan bir model yay\u0131nlayabilece\u011fini belirtti.<\/p>\n
Balderton Capital<\/b>‘de ortak olan James Wise<\/b>, 500 milyon sterlinlik h\u00fck\u00fcmet finansman\u0131 taraf\u0131ndan desteklenen \u0130ngiliz yapay zeka \u015firketlerine yat\u0131r\u0131m yapacak bir risk sermayesi fonu olan Sovereign AI<\/b> biriminin ba\u015fkan\u0131d\u0131r.<\/p>\n
Mythos<\/b>‘un sistemlerin g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 ortaya \u00e7\u0131karabilecek “daha bir\u00e7ok g\u00fc\u00e7l\u00fc modelin ilki” oldu\u011funu s\u00f6yledi.<\/p>\n
Birimi, BBC’nin Today Program\u0131na verdi\u011fi deme\u00e7te, “bunu ele alan \u0130ngiliz yapay zeka \u015firketlerine \u2013 yapay zeka g\u00fcvenli\u011fi ve emniyeti alan\u0131nda \u00e7al\u0131\u015fan \u015firketlere \u2013 yat\u0131r\u0131m yap\u0131yor.”<\/p>\n
“Umar\u0131z g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 ortaya \u00e7\u0131karan modeller, ayn\u0131 zamanda onlar\u0131 d\u00fczelten modeller de olur.”<\/p>\n
#YapayZeka #MythosAI #SiberG\u00fcvenlik #FinansSistemleri #Anthropic #ClaudeMythos #TeknolojiG\u00fcvenli\u011fi #AIEndi\u015feleri #FinansalRisk #DijitalG\u00fcvenlik<\/p>\n","protected":false},"excerpt":{"rendered":"
Finans bakanlar\u0131, merkez bankac\u0131lar\u0131 ve finans\u00f6rler, finansal sistemlerin g\u00fcvenli\u011fini zay\u0131flatabilece\u011finden korktuklar\u0131 g\u00fc\u00e7l\u00fc yeni bir yapay zeka modeli hakk\u0131nda ciddi endi\u015felerini dile getirdiler. Anthropic taraf\u0131ndan geli\u015ftirilen Claude Mythos modelinin, bir\u00e7ok b\u00fcy\u00fck i\u015fletim sistemindeki g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 bulmas\u0131n\u0131n ard\u0131ndan kriz toplant\u0131lar\u0131na yol a\u00e7t\u0131\u011f\u0131 belirtildi. Uzmanlar, modelin siber g\u00fcvenlik zay\u0131fl\u0131klar\u0131n\u0131 tespit etme ve kullanma konusunda e\u015fi benzeri g\u00f6r\u00fclmemi\u015f bir […]<\/p>\n","protected":false},"author":1,"featured_media":3292,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[40],"tags":[],"class_list":["post-3291","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-dunya-haberleri"],"_links":{"self":[{"href":"https:\/\/vanak.news\/index.php?rest_route=\/wp\/v2\/posts\/3291","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/vanak.news\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/vanak.news\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/vanak.news\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/vanak.news\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3291"}],"version-history":[{"count":0,"href":"https:\/\/vanak.news\/index.php?rest_route=\/wp\/v2\/posts\/3291\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/vanak.news\/index.php?rest_route=\/wp\/v2\/media\/3292"}],"wp:attachment":[{"href":"https:\/\/vanak.news\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3291"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/vanak.news\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3291"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/vanak.news\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3291"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}