{"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\/cs\/blog\/news\/anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej","title":{"rendered":"Anthropic varoval, \u017ee jeho budouc\u00ed model Claude Mythos se stane \"hackersk\u00fdm z\u00e1zrakem\".","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Firmy jako Anthropic, OpenAI a dal\u0161\u00ed technologick\u00e9 spole\u010dnosti vyv\u00edjej\u00ed AI modely, kter\u00e9 mohou v\u00fdrazn\u011b zv\u00fd\u0161it hrozbu masivn\u00edch kybernetick\u00fdch \u00fatok\u016f na firemn\u00ed, st\u00e1tn\u00ed a m\u011bstsk\u00e9 syst\u00e9my. V centru pozornosti je dosud nevydan\u00fd AI model Claude Mythos od spole\u010dnosti Anthropic. Podle informac\u00ed Axios firma ji\u017e varuje vysoce postaven\u00e9 p\u0159edstavitele USA, \u017ee jeho uveden\u00ed na trh v roce 2026 v\u00fdrazn\u011b zvy\u0161uje pravd\u011bpodobnost \u00fatok\u016f takov\u00e9ho rozsahu.<\/p>\n

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

Mythos je popisov\u00e1n jako syst\u00e9m, kter\u00fd umo\u017e\u0148uje AI agent\u016fm autonomn\u011b, s vysokou p\u0159esnost\u00ed a slo\u017eitost\u00ed, pronikat do firemn\u00edch, st\u00e1tn\u00edch a m\u011bstsk\u00fdch syst\u00e9m\u016f. \u0160\u00e9f Axiosu, Jim VandeHei, uvedl ve sv\u00e9m zpravodaji pro vedouc\u00ed pracovn\u00edky, \u017ee jeden ze zdroj\u016f, obezn\u00e1men\u00fdch s budouc\u00edmi AI modely, p\u0159ipou\u0161t\u00ed velk\u00fd kybernetick\u00fd \u00fatok u\u017e v roce 2026. Podniky jsou p\u0159itom ozna\u010deny za zvl\u00e1\u0161\u0165 zraniteln\u00e9 c\u00edle.<\/p>\n

D\u0159\u00edve Fortune zkoumal<\/a> \u00fanikov\u00e9 materi\u00e1ly z blogu Anthropic o Claude Mythos. Tento AI model je pops\u00e1n jako syst\u00e9m, kter\u00fd \"v sou\u010dasnosti v\u00fdrazn\u011b p\u0159ed\u010d\u00ed jak\u00fdkoli jin\u00fd AI model co se t\u00fd\u010de kybernetick\u00fdch schopnost\u00ed\". D\u00e1le se uv\u00e1d\u00ed, \u017ee Mythos \"p\u0159edzv\u011bst\u00ed nadch\u00e1zej\u00edc\u00ed vlnu AI model\u016f, schopn\u00fdch vyu\u017e\u00edvat zranitelnosti zp\u016fsobem, kter\u00fd v\u00fdrazn\u011b p\u0159ed\u010d\u00ed snahy o kybernetickou ochranu\". Riziko zvy\u0161uj\u00ed zam\u011bstnanci, kte\u0159\u00ed pou\u017e\u00edvaj\u00ed AI agenty bez pov\u011bdom\u00ed o tom, \u017ee to m\u016f\u017ee usnadnit kyberzlo\u010dinc\u016fm p\u0159\u00edstup k vnit\u0159n\u00edm syst\u00e9m\u016fm jejich firem.<\/p>\n

Na konci roku 2025 Anthropic odhalila<\/a> prvn\u00ed zdokumentovan\u00fd p\u0159\u00edpad kybernetick\u00e9ho \u00fatoku, kter\u00fd byl do zna\u010dn\u00e9 m\u00edry proveden AI. St\u00e1tem podporovan\u00e1 \u010d\u00ednsk\u00e1 skupina pou\u017eila AI agenty k autonomn\u00edmu prolomen\u00ed p\u0159ibli\u017en\u011b 30 c\u00edl\u016f po cel\u00e9m sv\u011bt\u011b. AI sama vedla 80 % a\u017e 90 % taktick\u00fdch operac\u00ed. K tomu do\u0161lo je\u0161t\u011b p\u0159ed n\u00e1hle v\u00fdrazn\u00fdm zv\u00fd\u0161en\u00edm agentn\u00edch syst\u00e9m\u016f a p\u0159ed vznikem nov\u00fdch nebezpe\u010dn\u00fdch zp\u016fsob\u016f jejich vyu\u017eit\u00ed.<\/p>\n

Nov\u00e9 AI modely se vyzna\u010duj\u00ed vy\u0161\u0161\u00ed schopnost\u00ed podporovat autonomn\u00ed pr\u00e1ci AI agent\u016f. Jde o syst\u00e9my, kter\u00e9 umo\u017e\u0148uj\u00ed agent\u016fm samostatn\u011b myslet, jednat, uva\u017eovat a improvizovat bez p\u0159est\u00e1vek a \u00fanavy. Rozsah \u00fatoku je nyn\u00ed ur\u010den p\u0159edev\u0161\u00edm v\u00fdpo\u010detn\u00edmi zdroji \u00fato\u010dn\u00edk\u016f, a ne po\u010dtem jejich skupin. Nyn\u00ed jeden \u010dlov\u011bk m\u016f\u017ee mo\u017enost<\/a> prov\u00e1d\u011bt kybernetick\u00e9 \u00fatoky, pro kter\u00e9 d\u0159\u00edve byly zapot\u0159eb\u00ed cel\u00e9 t\u00fdmy.<\/p>\n

Rostouc\u00ed kybernetick\u00e9 hrozby z\u00e1rove\u0148 zvy\u0161uj\u00ed zranitelnost spole\u010dnost\u00ed. Zam\u011bstnanci \u010d\u00edm d\u00e1l \u010dast\u011bji spou\u0161t\u00ed Claude, Copilot a dal\u0161\u00ed modely AI, \u010dasto z domova, a vytv\u00e1\u0159ej\u00ed sv\u00e9 vlastn\u00ed AI agenty. V mnoha p\u0159\u00edpadech maj\u00ed tito agenti bez \u0159\u00e1dn\u00e9ho dozoru p\u0159\u00edstup k intern\u00edm pracovn\u00edm syst\u00e9m\u016fm. To vytv\u00e1\u0159\u00ed pro kyberzlo\u010dince dal\u0161\u00ed kan\u00e1l pro pr\u016fnik. V odv\u011btv\u00ed se tento neautorizovan\u00fd nebo nekontrolovan\u00fd zp\u016fsob vyu\u017e\u00edv\u00e1n\u00ed AI naz\u00fdv\u00e1 \u201est\u00ednov\u00e1 AI\u201c.<\/p>\n

Pr\u016fzkum publikace Dark Reading uk\u00e1zal, \u017ee 48 % odborn\u00edk\u016f na kybernetickou bezpe\u010dnost pova\u017euje AI agenty za hlavn\u00ed sm\u011br \u00fatok\u016f v roce 2026. Toto nebezpe\u010d\u00ed je ozna\u010deno jako z\u00e1va\u017en\u011bj\u0161\u00ed ne\u017e deepfake a dal\u0161\u00ed rizika zahrnut\u00e1 v pr\u016fzkumu. Ka\u017ed\u00fd zam\u011bstnanec jak\u00e9koliv spole\u010dnosti na sv\u011bt\u011b by m\u011bl ji\u017e nyn\u00ed ch\u00e1pat riziko pou\u017e\u00edv\u00e1n\u00ed AI agent\u016f v bl\u00edzkosti citliv\u00fdch informac\u00ed. Technick\u00fd t\u00fdm Axios pova\u017euje toto nebezpe\u010d\u00ed za nejv\u011bt\u0161\u00ed sou\u010dasnou hrozbu pro spole\u010dnosti. Praktick\u00fd z\u00e1v\u011br je jasn\u00fd: k pr\u00e1ci s AI agenty je pot\u0159eba bezpe\u010dn\u00e9 izolovan\u00e9 prost\u0159ed\u00ed.<\/p>\n

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