{"id":166695,"date":"2026-03-30T11:13:41","date_gmt":"2026-03-30T09:13:49","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej"},"modified":"2026-03-31T11:20:25","modified_gmt":"2026-03-31T09:20:25","slug":"anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sk\/blog\/news\/anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej","title":{"rendered":"Anthropic varovala, \u017ee jej bud\u00faci model Claude Mythos sa stane \u201ehackerovou z\u00e1zra\u010dnou zbra\u0148ou\u201c","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Anthropic, OpenAI a \u010fal\u0161ie technologick\u00e9 firmy pripravuj\u00fa AI modely, ktor\u00e9 m\u00f4\u017eu dramaticky zv\u00fd\u0161i\u0165 hrozbu rozsiahlych kybernetick\u00fdch \u00fatokov na korpor\u00e1tne, \u0161t\u00e1tne a mestsk\u00e9 syst\u00e9my. V centre pozornosti sa nach\u00e1dza e\u0161te nevydan\u00fd AI model Claude Mythos od spolo\u010dnosti Anthropic. Pod\u013ea inform\u00e1ci\u00ed Axios spolo\u010dnos\u0165 u\u017e varuje vysokopostaven\u00fdch \u00faradn\u00edkov USA, \u017ee jeho pr\u00edchod v roku 2026 v\u00fdrazne zvy\u0161uje pravdepodobnos\u0165 \u00fatokov tohto druhu.<\/p>\n

<\/center>\"Anthropic<\/p>\n

Mythos je popisovan\u00fd ako syst\u00e9m, ktor\u00fd umo\u017e\u0148uje AI agentom auton\u00f3mne, s vysokou presnos\u0165ou a zlo\u017eit\u00fdm sp\u00f4sobom prenikn\u00fa\u0165 do korpor\u00e1tnych, \u0161t\u00e1tnych a mestsk\u00fdch syst\u00e9mov. Gener\u00e1lny riadite\u013e Axios Jim VandeHei vo svojej spr\u00e1ve pre ved\u00facich uviedol, \u017ee jeden zo zdrojov obozn\u00e1men\u00fdch s bud\u00facimi AI modelmi predpoklad\u00e1 ve\u013ek\u00fd kybernetick\u00fd \u00fatok u\u017e v roku 2026. Pri tom bol podnik ozna\u010den\u00fd za obzvl\u00e1\u0161\u0165 zranite\u013en\u00fd cie\u013e.<\/p>\n

Ned\u00e1vno Fortune presk\u00famal<\/a> uniknut\u00e9 materi\u00e1ly z blogu Anthropic o Claude Mythos. V \u0148om je AI model nazvan\u00fd syst\u00e9mom, ktor\u00fd \u201ev s\u00fa\u010dasnosti v\u00fdrazne prekon\u00e1va ak\u00fdko\u013evek in\u00fd AI model z h\u013eadiska kybernetick\u00fdch schopnost\u00ed\u201c. V \u0148om sa tie\u017e uv\u00e1dza, \u017ee Mythos \u201epredpoved\u00e1 prich\u00e1dzaj\u00facu vlnu AI modelov, schopn\u00fdch vyu\u017e\u00edva\u0165 zranite\u013enosti tak, aby to v\u00fdrazne predbehlo snahy o kybernetick\u00fa obranu\u201c. Riziko zvy\u0161uj\u00fa zamestnanci, ktor\u00ed pou\u017e\u00edvaj\u00fa AI agentov bez porozumenia tomu, \u017ee to m\u00f4\u017ee u\u013eah\u010di\u0165 kyberzlo\u010dincom pr\u00edstup k intern\u00fdm syst\u00e9mom ich spolo\u010dnost\u00ed.<\/p>\n

Na konci roku 2025 Anthropic odhalila<\/a> prv\u00fd zdokumentovan\u00fd pr\u00edpad kybernetick\u00e9ho \u00fatoku, do zna\u010dnej miery vykonan\u00e9ho AI. \u0160t\u00e1tnou podporovan\u00e1 \u010d\u00ednska skupina pou\u017eila AI agentov na auton\u00f3mne hackovanie pribli\u017ene 30 cie\u013eov po celom svete. AI samo vykon\u00e1valo od 80 % do 90 % taktick\u00fdch oper\u00e1ci\u00ed. To sa stalo e\u0161te pred prudk\u00fdm posilnen\u00edm agenty syst\u00e9mov a pred vznikom nov\u00fdch nebezpe\u010dn\u00fdch sp\u00f4sobov ich vyu\u017eitia.<\/p>\n

Nov\u00e9 AI modely sa vyzna\u010duj\u00fa vy\u0161\u0161ou schopnos\u0165ou podporova\u0165 auton\u00f3mnu \u010dinnos\u0165 AI agentov. Ide o syst\u00e9my, ktor\u00e9 umo\u017e\u0148uj\u00fa agentom samostatne myslie\u0165, kona\u0165, usudzova\u0165 a improvizova\u0165 bez prest\u00e1vok a \u00fanavy. Rozsah \u00fatoku je teraz definovan\u00fd predov\u0161etk\u00fdm v\u00fdpo\u010dtov\u00fdmi zdrojmi \u00fato\u010dn\u00edkov, a nie po\u010dtom ich skup\u00edn. Teraz jeden \u010dlovek dost\u00e1va mo\u017enos\u0165<\/a> vykon\u00e1vanie kybernetick\u00fdch \u00fatokov, na ktor\u00e9 predt\u00fdm potrebn\u00e9 cel\u00e9 t\u00edmy.<\/p>\n

S\u00fa\u010dasne rastie aj zranite\u013enos\u0165 spolo\u010dnost\u00ed. Zamestnanci \u010doraz \u010dastej\u0161ie sp\u00fa\u0161\u0165aj\u00fa Claude, Copilot a in\u00e9 agentov\u00e9 modely, \u010dasto z domu, a vytv\u00e1raj\u00fa si vlastn\u00fdch AI agentov. V mnoh\u00fdch pr\u00edpadoch tak\u00e9to agenti bez riadneho doh\u013eadu z\u00edskavaj\u00fa pr\u00edstup k intern\u00fdm pracovn\u00fdm syst\u00e9mom. To vytv\u00e1ra pre kyberzlo\u010dincov e\u0161te jeden kan\u00e1l prenikania. V odvetv\u00ed sa tak\u00e9mu neopr\u00e1vnen\u00e9mu alebo nekontrolovan\u00e9mu pou\u017e\u00edvaniu AI hovor\u00ed \u201etie\u0148ov\u00fd AI\u201c.<\/p>\n

Prieskum zdroja Dark Reading uk\u00e1zal, \u017ee 48 % odborn\u00edkov na kybernetick\u00fa bezpe\u010dnos\u0165 pova\u017euje agentn\u00fd AI za hlavn\u00fa hrozbu \u00fatokov v roku 2026. T\u00e1to hrozba bola hodnoten\u00e1 vy\u0161\u0161ie ne\u017e deepfakes a v\u0161etky ostatn\u00e9 rizik\u00e1 zahrnut\u00e9 v prieskume. Ka\u017ed\u00fd zamestnanec akejko\u013evek spolo\u010dnosti na svete by u\u017e teraz mal ch\u00e1pa\u0165 riziko pou\u017e\u00edvania AI agentov bl\u00edzko citlivej inform\u00e1cie. Technick\u00fd t\u00edm Axios pova\u017euje t\u00fato hrozbu za najv\u00e4\u010d\u0161iu pre spolo\u010dnos\u0165 v s\u00fa\u010dasnosti. Praktick\u00fd z\u00e1ver je jasn\u00fd: na pr\u00e1cu s AI agentmi je potrebn\u00e9 bezpe\u010dn\u00e9 izolovan\u00e9 prostredie.<\/p>\n

Zdroj:<\/strong><\/p>\n