{"id":167916,"date":"2026-04-08T11:12:35","date_gmt":"2026-04-08T09:12:39","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/anthropic-otkryla-ogranichennyj-dostup-k-modeli-claude-mythos-preview-ona-avtonomno-ishhet-dyry-v-po-i-uzhe-nashla-tysyachi-uyazvimostej"},"modified":"2026-04-10T10:23:35","modified_gmt":"2026-04-10T08:23:35","slug":"anthropic-otkryla-ogranichennyj-dostup-k-modeli-claude-mythos-preview-ona-avtonomno-ishhet-dyry-v-po-i-uzhe-nashla-tysyachi-uyazvimostej","status":"publish","type":"post","link":"https:\/\/prohoster.info\/cs\/blog\/news\/anthropic-otkryla-ogranichennyj-dostup-k-modeli-claude-mythos-preview-ona-avtonomno-ishhet-dyry-v-po-i-uzhe-nashla-tysyachi-uyazvimostej","title":{"rendered":"Anthropic otev\u0159el omezen\u00fd p\u0159\u00edstup k modelu Claude Mythos Preview \u2014 autonomn\u011b hled\u00e1 slabiny v softwaru a ji\u017e na\u0161el tis\u00edce zranitelnost\u00ed.","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Spole\u010dnost Anthropic otev\u0159ela omezen\u00fd p\u0159\u00edstup k sv\u00e9mu nov\u00e9mu AI modelu Mythos, kter\u00fd je ur\u010den pro hled\u00e1n\u00ed kritick\u00fdch zranitelnost\u00ed a odr\u00e1\u017een\u00ed kybernetick\u00fdch hrozeb. N\u00e1stroj bude dostupn\u00fd pouze vybran\u00fdm korporac\u00edm, proto\u017ee jeho schopnosti hackov\u00e1n\u00ed syst\u00e9m\u016f byly pova\u017eov\u00e1ny za p\u0159\u00edli\u0161 nebezpe\u010dn\u00e9 pro ve\u0159ejn\u00e9 vyd\u00e1n\u00ed.<\/p>\n

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

Podle zpr\u00e1v\u011b<\/a> The Wall Street Journal, p\u0159edb\u011b\u017enou verzi neuronov\u00e9 s\u00edt\u011b obdr\u017e\u00ed asi 50 spole\u010dnost\u00ed a organizac\u00ed podporuj\u00edc\u00edch kritickou IT infrastrukturu, mezi nimi\u017e jsou Amazon, Apple, Google a Linux Foundation. Podle vedouc\u00edho odd\u011blen\u00ed Frontier Red Team ve spole\u010dnosti Anthropic Logana Grahama, AI model Mythos prok\u00e1zal natolik vysok\u00e9 schopnosti v hled\u00e1n\u00ed a vyu\u017e\u00edv\u00e1n\u00ed softwarov\u00fdch chyb, \u017ee v\u00fdvoj\u00e1\u0159i v sou\u010dasnosti nepl\u00e1nuj\u00ed jej otev\u0159\u00edt \u0161irok\u00e9 ve\u0159ejnosti.<\/p>\n

Odborn\u00edci na kybernetickou bezpe\u010dnost jsou \u010d\u00edm d\u00e1l t\u00edm v\u00edce znepokojeni t\u00edm, \u017ee um\u011bl\u00e1 inteligence rychle zkracuje dobu mezi t\u00edm, kdy je zranitelnost odhalena, a t\u00edm, kdy je vyvinut funk\u010dn\u00ed software pro jej\u00ed vyu\u017eit\u00ed. Na konci lo\u0148sk\u00e9ho roku zjistili v\u00fdzkumn\u00edci na Stanfordsk\u00e9 univerzit\u011b, \u017ee neuronov\u00e9 s\u00edt\u011b hledaj\u00ed a vyu\u017e\u00edvaj\u00ed chyby v re\u00e1ln\u00fdch s\u00edt\u00edch t\u00e9m\u011b\u0159 stejn\u011b efektivn\u011b jako lid\u0161t\u00ed odborn\u00edci. P\u0159\u00edkladem je p\u0159edchoz\u00ed model Anthropic, Claude Opus 4.6, kter\u00fd na za\u010d\u00e1tku tohoto roku b\u011bhem pouh\u00fdch dvou t\u00fddn\u016f na\u0161el v prohl\u00ed\u017ee\u010di Firefox v\u00edce kritick\u00fdch zranitelnost\u00ed, ne\u017e obvykle odhal\u00ed cel\u00fd sv\u011bt za dva m\u011bs\u00edce.<\/p>\n

Nov\u00fd model Mythos se uk\u00e1zal b\u00fdt p\u0159ibli\u017en\u011b 10kr\u00e1t efektivn\u011bj\u0161\u00ed ne\u017e p\u0159edchoz\u00ed AI syst\u00e9my p\u0159i v\u00fdpo\u010dtu n\u00e1klad\u016f na odhalen\u00ed jedn\u00e9 chyby. I kdy\u017e Anthropic zat\u00edm nepl\u00e1nuje uvolnit sv\u016fj produkt pro ve\u0159ejnost, Logan Graham je p\u0159esv\u011bd\u010den, \u017ee v p\u0159\u00ed\u0161t\u00edch n\u011bkolika letech se podobn\u00e9 hackersk\u00e9 schopnosti objev\u00ed i u dal\u0161\u00edch neuronov\u00fdch s\u00edt\u00ed. V t\u00e9to souvislosti odborn\u00edk vyz\u00fdv\u00e1 IT komunitu, aby se ji\u017e nyn\u00ed p\u0159ipravila na realitu, ve kter\u00e9 bude doba mezi odhalen\u00edm softwarov\u00e9 zranitelnosti a jej\u00edm pou\u017eit\u00edm v \u00fatoku zkr\u00e1cena na nulu.<\/p>\n

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