Nov\u00fd AI model Mythos dok\u00e1\u017ee v \u0159\u00e1du minut zanalyzovat miliony \u0159\u00e1dk\u016f k\u00f3du a naj\u00edt cestu i do nejst\u0159e\u017een\u011bj\u0161\u00edch s\u00edt\u00ed. Asymetrie v p\u0159\u00edstupu k technologii, kterou USA dr\u017e\u00ed pod pokli\u010dkou a Evrop\u011b ji zat\u00edm odp\u00edraj\u00ed, vyvol\u00e1v\u00e1 bezpe\u010dnostn\u00ed vakuum a nap\u011bt\u00ed.\u00a0<\/p>\n
Evropsk\u00e9 politick\u00e9 i finan\u010dn\u00ed \u0161pi\u010dky zachv\u00e1tila panika. Zat\u00edmco se za Atlantikem buduj\u00ed obrann\u00e9 valy, Evropa z\u016fst\u00e1v\u00e1 od testov\u00e1n\u00ed nebezpe\u010dn\u011b v\u00fdkonn\u00e9ho, a proto zat\u00edm\u00a0utajovan\u00e9ho<\/a>\u00a0AI modelu Mythos nad\u00e1le od\u0159\u00edznuta. Evropsk\u00e1 digit\u00e1ln\u00ed infrastruktura v\u010detn\u011b finan\u010dn\u00edho syst\u00e9mu se tak m\u016f\u017ee st\u00e1t snadn\u00fdm ter\u010dem pro kohokoli, komu se tato nov\u00e1 digit\u00e1ln\u00ed superzbra\u0148 dostane do rukou.\u00a0<\/p>\n Znepokojiv\u00e9 zpr\u00e1vy o schopnostech modelu, p\u0159edstaven\u00e9ho americkou spole\u010dnost\u00ed\u00a0Anthropic<\/a>\u00a0p\u0159ed m\u011bs\u00edcem, o\u017eivily nejhor\u0161\u00ed no\u010dn\u00ed m\u016fry mana\u017eer\u016f: v \u00e9\u0159e AI m\u016f\u017ee katastrofa pro byznys p\u0159ij\u00edt odkudkoli, ne\u010dekan\u011b a s drtivou rychlost\u00ed. Mythos toti\u017e nen\u00ed jen dal\u0161\u00edm chatbotem pro generov\u00e1n\u00ed text\u016f, ale doslova pred\u00e1torem schopn\u00fdm vyhled\u00e1vat dosud nezn\u00e1m\u00e9 chyby v samotn\u00e9m j\u00e1dru opera\u010dn\u00edch syst\u00e9m\u016f a prohl\u00ed\u017ee\u010d\u016f. M\u016f\u017ee se tak snadno st\u00e1t n\u00e1strojem kyberzlo\u010dinc\u016f, kte\u0159\u00ed pronikaj\u00ed do digit\u00e1ln\u00edch syst\u00e9m\u016f, aby je ovl\u00e1dli.\u00a0<\/p>\n Nap\u011bt\u00ed kolem Mythosu umoc\u0148uje pr\u00e1v\u011b extr\u00e9mn\u011b nerovnom\u011brn\u00fd p\u0159\u00edstup k modelu. Zat\u00edmco Amerika svou infrastrukturu na n\u00e1stup tak mocn\u00e9ho a zneu\u017eiteln\u00e9ho n\u00e1stroje u\u017e aktivn\u011b p\u0159ipravuje, zbytek sv\u011bta v\u010detn\u011b Evropy nech\u00e1v\u00e1 \u010dekat v p\u0159eds\u00e1l\u00ed.\u00a0<\/p>\n V pond\u011bl\u00ed proto Evropsk\u00e1 komise\u00a0vstoupila do jedn\u00e1n\u00ed<\/a>\u00a0s Anthropicem, aby umo\u017enil testov\u00e1n\u00ed modelu i evropsk\u00fdmi spole\u010dnostmi v souladu s evropsk\u00fdm kodexem pro AI. Do\u0161lo k tomu na pozad\u00ed jedn\u00e1n\u00ed ministr\u016f financ\u00ed euroz\u00f3ny svolan\u00e9ho pr\u00e1v\u011b kv\u016fli rizik\u016fm, je\u017e Mythos p\u0159edstavuje pro evropskou digit\u00e1ln\u00ed infrastrukturu a finan\u010dn\u00ed syst\u00e9m.\u00a0<\/p>\n Evrop\u0161t\u00ed p\u0159edstavitel\u00e9, v\u010detn\u011b Christine Lagarde z European Central Bank, u\u017e d\u0159\u00edve varovali, \u017ee bez p\u0159\u00edstupu k tomuto modelu nemohou evropsk\u00e9 banky a kritick\u00e1 infrastruktura testovat svou odolnost a p\u0159izp\u016fsobit se nov\u00fdm hrozb\u00e1m. \u017de tyto obavy nejsou plan\u00e9, dokl\u00e1d\u00e1 zpr\u00e1va, \u017ee se p\u0159es extern\u00edho dodavatele k modelu\u00a0dostala<\/a>\u00a0skupina neautorizovan\u00fdch u\u017eivatel\u016f.\u00a0<\/p>\n Bolestn\u00e1 asymetrie<\/p>\n P\u0159esto\u017ee evropsk\u00e9 banky m\u011bly podle d\u0159\u00edv\u011bj\u0161\u00edch informac\u00ed agentury Reuters\u00a0z\u00edskat p\u0159\u00edstup<\/a>\u00a0k Mythosu v \u0159\u00e1du dn\u016f \u010di t\u00fddn\u016f, spolu s vl\u00e1dami a regul\u00e1tory i nad\u00e1le klepou na dve\u0159e marn\u011b. Evropsk\u00e1 centr\u00e1ln\u00ed banka u\u017e proto kv\u016fli Mythosu svolala<\/a>\u00a0mimo\u0159\u00e1dn\u00e1 jedn\u00e1n\u00ed s \u0159editeli rizik nejv\u011bt\u0161\u00edch bankovn\u00edch dom\u016f v euroz\u00f3n\u011b a n\u011bmeck\u00e1 centr\u00e1ln\u00ed banka Deutsche Bundesbank spolu s n\u011bmeck\u00fdm \u00fa\u0159adem pro dohled nad finan\u010dn\u00edm trhem BaFin instituc\u00edm na\u0159\u00eddily<\/a>,\u00a0aby se okam\u017eit\u011b p\u0159ipravily na rychl\u00e9 zav\u00e1d\u011bn\u00ed softwarov\u00fdch z\u00e1plat.\u00a0\u00a0<\/p>\n V Americe p\u0159itom Mythos v r\u00e1mci elitn\u00edho Projektu Glasswing testuj\u00ed ameri\u010dt\u00ed giganti jako Apple, Google \u010di nejv\u011bt\u0161\u00ed banka JPMorgan Chase u\u017e t\u00fddny. Americk\u00e9 \u00fa\u0159ady vyu\u017e\u00edvaj\u00ed Mythos k prov\u011b\u0159en\u00ed odolnosti sv\u00fdch s\u00edt\u00ed, v\u010detn\u011b ministerstva financ\u00ed, Feder\u00e1ln\u00edho rezervn\u00edho syst\u00e9mu (Fed) nebo N\u00e1rodn\u00ed bezpe\u010dnostn\u00ed agentury (NSA).\u00a0<\/p>\n Obdobn\u011b si po\u010d\u00ednaj\u00ed tak\u00e9 \u00fa\u010dastn\u00edci testov\u00e1n\u00ed ze soukrom\u00e9ho sektoru. Telekomunika\u010dn\u00ed skupina Cisco Systems nasadila Mythos na vlastn\u00ed infrastrukturu, aby prov\u011b\u0159ila zranitelnost sv\u00fdch s\u00ed\u0165ov\u00fdch za\u0159\u00edzen\u00ed po cel\u00e9m sv\u011bt\u011b. \u201eM\u00e1m pocit, \u017ee bude existovat sv\u011bt p\u0159ed Mythosem a sv\u011bt po n\u011bm,\u201c\u00a0prohl\u00e1sil<\/a>\u00a0podle agentury Bloomberg Jeetu Patel, prezident a produktov\u00fd \u0159editel spole\u010dnosti.\u00a0<\/p>\n Zd\u00e1 se, \u017ee tato asymetrie mezi USA a zbytkem sv\u011bta bude pokra\u010dovat. B\u00edl\u00fd d\u016fm v \u00fater\u00fd\u00a0ozn\u00e1mil<\/a>,\u00a0\u017ee se spole\u010dnosti Google, Microsoft a xAI p\u0159ipoj\u00ed k firm\u00e1m OpenAI a Anthropic, aby americk\u00e9 vl\u00e1d\u011b poskytovaly p\u0159ednostn\u00ed p\u0159\u00edstup ke sv\u00fdm model\u016fm um\u011bl\u00e9 inteligence je\u0161t\u011b p\u0159ed jejich ve\u0159ejn\u00fdm p\u0159edstaven\u00edm.\u00a0<\/p>\n Asijsk\u00e9 trhy v prvn\u00ed linii<\/p>\n \u00dadajn\u00e9 schopnosti Mythosu se u\u017e v polovin\u011b dubna staly hlavn\u00edm t\u00e9matem jarn\u00edho zased\u00e1n\u00ed Mezin\u00e1rodn\u00edho m\u011bnov\u00e9ho fondu (IMF) ve Washingtonu. Tam dokonce zast\u00ednily t\u00e9ma ekonomick\u00fdch dopad\u016f v\u00e1lky na Bl\u00edzk\u00e9m v\u00fdchod\u011b a uzav\u0159en\u00ed Hormuzsk\u00e9ho pr\u016flivu, kter\u00e9 blokuje sv\u011btov\u00e9 dod\u00e1vky paliv a dal\u0161\u00edch komodit. \u201eRozd\u00edl oproti vojensk\u00e9mu nap\u011bt\u00ed v Hormuzsk\u00e9m pr\u016flivu je ten, \u017ee tam v\u00edme, kde hrozba le\u017e\u00ed a jak je velk\u00e1, zat\u00edmco u Anthropicu m\u00e1me co d\u011blat s n\u011b\u010d\u00edm naprosto nezn\u00e1m\u00fdm,\u201c popsal situaci kolem Mythosu kanadsk\u00fd ministr financ\u00ed Fran\u00e7ois-Philippe Champagne.\u00a0<\/p>\n Nedostatek p\u0159\u00edm\u00e9ho p\u0159\u00edstupu k modelu\u00a0vyvol\u00e1v\u00e1<\/a>\u00a0nejistotu tak\u00e9 na dal\u0161\u00edch trz\u00edch. Za\u0159azen\u00ed na seznam opr\u00e1vn\u011bn\u00fdch u\u017eivatel\u016f se proto u Anthropicu dom\u00e1haj\u00ed nap\u0159\u00edklad indick\u00e9 platebn\u00ed startupy a finan\u010dn\u011b-technologick\u00e9 platformy jako Razorpay, kter\u00e9 ka\u017edodenn\u011b zprost\u0159edkov\u00e1vaj\u00ed miliony transakc\u00ed.\u00a0<\/p>\n Kybernetick\u00e1 hrozba\u00a0vyd\u011bsila<\/a>\u00a0\u00fa\u0159ady v Asii, je\u017e plo\u0161n\u011b stup\u0148uj\u00ed dohled nad finan\u010dn\u00edm sektorem. Singapursk\u00fd m\u011bnov\u00fd \u00fa\u0159ad ve spolupr\u00e1ci s n\u00e1rodn\u00ed kybernetickou agenturou na\u0159\u00eddil bank\u00e1m okam\u017eit\u00e9 prov\u011b\u0159en\u00ed obrany a v\u010dasn\u00e9 z\u00e1platov\u00e1n\u00ed syst\u00e9m\u016f.\u00a0<\/p>\n Australsk\u00fd regul\u00e1tor finan\u010dn\u00edho syst\u00e9mu\u00a0upozornil<\/a>\u00a0tamn\u00ed banky, \u017ee jejich bezpe\u010dnostn\u00ed procesy nest\u00edhaj\u00ed s AI dr\u017eet krok. P\u0159edseda bezpe\u010dnostn\u00edho v\u00fdboru jihokorejsk\u00e9 n\u00e1rodn\u00ed rady pro AI I W\u014fn-tche situaci popsal lapid\u00e1rn\u011b: \u201eVstoupili jsme do \u00e9ry, kdy iniciativu v hackov\u00e1n\u00ed dr\u017e\u00ed um\u011bl\u00e1 inteligence, a nikoli lid\u00e9.\u201c\u00a0<\/p>\n Digit\u00e1ln\u00ed zlod\u011bj, kter\u00fd ne\u010dek\u00e1<\/p>\n Nervozita kolem nejpokro\u010dilej\u0161\u00edho modelu Anthropicu je odrazem obav, je\u017e bud\u00ed \u0161\u00e9fy firem ze span\u00ed st\u00e1le \u010dast\u011bji. Skokov\u00fd v\u00fdvoj um\u011bl\u00e9 inteligence znamen\u00e1 nejen siln\u011bj\u0161\u00ed n\u00e1stroj pro pos\u00edlen\u00ed kybernetick\u00e9 bezpe\u010dnosti, ale i to, \u017ee jej mohou vyu\u017e\u00edt hacke\u0159i pro hled\u00e1n\u00ed d\u011br v dosavadn\u00ed obran\u011b. \u010cas, kter\u00fd dosud instituce m\u011bly na opravu chyb ve sv\u00fdch softwarech, se tak dramaticky zkr\u00e1t\u00ed.\u00a0<\/p>\n Dosavadn\u00ed praxe funguje tak, \u017ee kdy\u017e tv\u016frce softwaru objev\u00ed v programu bezpe\u010dnostn\u00ed chybu, upozorn\u00ed na ni u\u017eivatele a z\u00e1kazn\u00edky \u2013 t\u0159eba banky nebo \u00fa\u0159ady. Ti pak maj\u00ed t\u00fddny na nasazen\u00ed pot\u0159ebn\u00fdch bezpe\u010dnostn\u00edch z\u00e1plat. Mythos tento proces naru\u0161uje \u2013 nevytv\u00e1\u0159\u00ed sice \u017e\u00e1dn\u00e9 nov\u00e9 viry, ale um\u00ed bleskov\u011b proch\u00e1zet miliony \u0159\u00e1dk\u016f st\u00e1vaj\u00edc\u00edho k\u00f3du a vyhled\u00e1vat v n\u011bm drobn\u00e9 chyby zanechan\u00e9 program\u00e1tory. Tyto izolovan\u00e9 defekty model zanalyzuje a dok\u00e1\u017ee je vyu\u017e\u00edt k z\u00edsk\u00e1n\u00ed kontroly nad syst\u00e9mem. A to d\u0159\u00edv, ne\u017e sta\u010d\u00ed zareagovat \u010dlov\u011bk.\u00a0\u00a0<\/p>\n Lze to jednodu\u0161e popsat analogi\u00ed s bytov\u00fdm zlod\u011bjem: jedna chyba v k\u00f3du funguje jako pootev\u0159en\u00e9 okno, kter\u00fdm se model dostane do syst\u00e9mu, pot\u00e9 najde druhou chybu k odemknut\u00ed pomysln\u00fdch hlavn\u00edch dve\u0159\u00ed a pomoc\u00ed t\u0159et\u00ed chyby nakonec vy\u0159ad\u00ed alarm. Jednotliv\u00e9 kroky samy o sob\u011b nep\u0159edstavuj\u00ed p\u0159\u00edm\u00e9 ohro\u017een\u00ed, ale dohromady v rychl\u00e9m sledu poskytnou \u00fato\u010dn\u00edkovi pln\u00fd p\u0159\u00edstup k s\u00edti. Finan\u010dn\u00ed sv\u011bt, historicky postaven\u00fd na rigidn\u00edch a zdlouhav\u00fdch procesech testov\u00e1n\u00ed IT syst\u00e9m\u016f, t\u00edm ztr\u00e1c\u00ed schopnost v\u010das reagovat.\u00a0<\/p>\n Tento posun jasn\u011b ilustruj\u00ed oborov\u00e1 data. Zat\u00edmco podle specializovan\u00e9 platformy\u00a0zerodayclock.com<\/a>\u00a0trvalo v roce 2018 v pr\u016fm\u011bru 771 dn\u00ed, ne\u017e \u00fato\u010dn\u00edci dok\u00e1zali prom\u011bnit ve\u0159ejn\u011b ozn\u00e1menou softwarovou trhlinu ve funk\u010dn\u00ed n\u00e1stroj pro kybernetick\u00fd \u00fatok, dnes se tato doba zkr\u00e1tila na m\u00e9n\u011b ne\u017e \u010dty\u0159i hodiny. To p\u0159edstavuje hrozbu zejm\u00e9na pro velk\u00e9 banky, proto\u017ee z\u00e1sah do kritick\u00fdch syst\u00e9m\u016f vy\u017eaduje v\u00edcestup\u0148ov\u00e9 schvalov\u00e1n\u00ed.\u00a0<\/p>\n Ti, kdo u\u017e maj\u00ed model k dispozici, varuj\u00ed p\u0159ed bagatelizac\u00ed situace. Podle Bloombergu\u00a0provedl<\/a>\u00a0dosud jedin\u00fd nez\u00e1visl\u00fd audit Mythosu britsk\u00fd Institut pro bezpe\u010dnost AI (AISI). Jeho analytici zjistili<\/a>,\u00a0\u017ee model skute\u010dn\u011b objevuje trhliny a prov\u00e1d\u00ed v\u00edcestup\u0148ov\u00e9 \u00fatoky autonomn\u011b, z\u00e1rove\u0148 v\u0161ak upozornili, \u017ee nejv\u00edce ohro\u017euje star\u0161\u00ed a zraniteln\u00e9 syst\u00e9my.\u00a0<\/p>\n Glob\u00e1ln\u00ed konglomer\u00e1ty si \u0161pi\u010dkovou obranu IT syst\u00e9m\u016f d\u0159\u00edve \u010di pozd\u011bji nakoup\u00ed. Hlavn\u00ed ob\u011bt\u00ed nov\u00e9 kybernetick\u00e9 reality se tak pravd\u011bpodobn\u011b stanou st\u0159edn\u00ed podniky a lok\u00e1ln\u00ed instituce, pro kter\u00e9 bude udr\u017een\u00ed kroku s autonomn\u00edmi \u00fatoky person\u00e1ln\u011b i finan\u010dn\u011b mimo\u0159\u00e1dn\u011b n\u00e1ro\u010dn\u00e9.\u00a0<\/p>\n","protected":false},"excerpt":{"rendered":"Nov\u00fd AI model Mythos dok\u00e1\u017ee v \u0159\u00e1du minut zanalyzovat miliony \u0159\u00e1dk\u016f k\u00f3du a naj\u00edt cestu i do nejst\u0159e\u017een\u011bj\u0161\u00edch…\n","protected":false},"author":2,"featured_media":79971,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[5857,11611,109,108,7968,106,321,8645,466,25558,845,113],"class_list":{"0":"post-79970","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-evropa","8":"tag-anthropic","9":"tag-banka","10":"tag-europa","11":"tag-europe","12":"tag-eurozona","13":"tag-evropa","14":"tag-evropska-centralni-banka","15":"tag-federalni-rezervni-system","16":"tag-google","17":"tag-mythos","18":"tag-umela-inteligence","19":"tag-usa"},"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@cz\/116542573432711012","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/79970","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/comments?post=79970"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/79970\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media\/79971"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media?parent=79970"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/categories?post=79970"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/tags?post=79970"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}