{"id":170327,"date":"2026-08-06T09:17:12","date_gmt":"2026-08-06T09:17:12","guid":{"rendered":"https:\/\/www.europesays.com\/cz\/170327\/"},"modified":"2026-08-06T09:17:12","modified_gmt":"2026-08-06T09:17:12","slug":"problemy-ma-uz-i-meta-take-jeji-ai-agent-utocil-jako-hacker","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/cz\/170327\/","title":{"rendered":"Probl\u00e9my m\u00e1 u\u017e i\u00a0Meta. Tak\u00e9 jej\u00ed AI agent \u00fato\u010dil jako hacker"},"content":{"rendered":"

P\u0159\u00edpad je dal\u0161\u00edm d\u016fkazem, \u017ee nejde o\u00a0ojedin\u011bl\u00e9 selh\u00e1n\u00ed jedin\u00e9 firmy. B\u011bhem kr\u00e1tk\u00e9 doby u\u017e podobn\u00e9 incidenty potvrdili t\u0159i nejv\u011bt\u0161\u00ed v\u00fdvoj\u00e1\u0159i pokro\u010dil\u00e9 um\u011bl\u00e9 inteligence na sv\u011bt\u011b.<\/p>\n

Meta uvedla, \u017ee probl\u00e9m vznikl b\u011bhem bezpe\u010dnostn\u00edho testov\u00e1n\u00ed jednoho z\u00a0jej\u00edch AI model\u016f. Kv\u016fli chyb\u011b v\u00a0konfiguraci testovac\u00edho prost\u0159ed\u00ed z\u00edskal p\u0159\u00edstup na ve\u0159ejn\u00fd internet, p\u0159esto\u017ee jej m\u00edt nem\u011bl. N\u00e1sledn\u011b pronikl do syst\u00e9mu jin\u00e9 organizace.<\/p>\n

Spole\u010dnost zat\u00edm neup\u0159esnila, o\u00a0jakou firmu \u0161lo, ani jak\u00e9 n\u00e1sledky m\u011bl cel\u00fd incident. Zd\u016fraznila pouze, \u017ee p\u0159\u00edpad st\u00e1le vy\u0161et\u0159uje a\u00a0podrobn\u011bj\u0161\u00ed informace zve\u0159ejn\u00ed, a\u017e bude m\u00edt k\u00a0dispozici v\u0161echny skute\u010dnosti.<\/p>\n

<\/a>Stejn\u00fd sc\u00e9n\u00e1\u0159 jako u\u00a0konkurence<\/p>\n

Na cel\u00e9 ud\u00e1losti je zaj\u00edmav\u00e9, \u017ee bezpe\u010dnostn\u00ed testy pro Metu prov\u00e1d\u011bla spole\u010dnost Irregular. Stejn\u00e1 firma zaji\u0161\u0165ovala tak\u00e9 testov\u00e1n\u00ed model\u016f Anthropicu.<\/p>\n

Podle jej\u00edho vyj\u00e1d\u0159en\u00ed byl incident prakticky toto\u017en\u00fd s\u00a0t\u00edm, kter\u00fd Anthropic zve\u0159ejnil minul\u00fd t\u00fdden. I\u00a0tehdy se model dostal na ve\u0159ejn\u00fd internet kv\u016fli chybn\u011b nastaven\u00e9mu testovac\u00edmu prost\u0159ed\u00ed.<\/p>\n

Anthropic n\u00e1sledn\u011b p\u0159i zp\u011btn\u00e9 kontrole objevil t\u0159i p\u0159\u00edpady, kdy jeho modely pronikly do syst\u00e9m\u016f r\u016fzn\u00fdch organizac\u00ed. Je\u0161t\u011b p\u0159edt\u00edm OpenAI p\u0159iznala, \u017ee jej\u00ed experiment\u00e1ln\u00ed AI agent b\u011bhem intern\u00edho testov\u00e1n\u00ed napadl platformu Hugging Face, vyu\u017eil n\u011bkolik dal\u0161\u00edch internetov\u00fdch slu\u017eeb a\u00a0b\u011bhem \u010dty\u0159 dn\u016f provedl v\u00edce ne\u017e 17\u00a0tis\u00edc r\u016fzn\u00fdch akc\u00ed.<\/p>\n

AI hledala cestu k\u00a0c\u00edli<\/p>\n

P\u0159esto\u017ee v\u0161echny t\u0159i p\u0159\u00edpady vypadaj\u00ed jako hackersk\u00e9 \u00fatoky, odborn\u00edci upozor\u0148uj\u00ed, \u017ee um\u011bl\u00e1 inteligence nejednala ze zl\u00e9 v\u016fle. \u201eNejsou v\u011bdom\u00e9 a\u00a0z\u00e1m\u011brn\u011b ned\u011blaj\u00ed nic z\u00e1ke\u0159n\u00e9ho,\u201c \u0159ekl BBC<\/a> Daniel Hulme, glob\u00e1ln\u00ed \u0159editel pro um\u011blou inteligenci spole\u010dnosti WPP.<\/p>\n

Podle n\u011bj modely pouze hledaj\u00ed co nej\u00fa\u010dinn\u011bj\u0161\u00ed strategii, jak splnit \u00fakol, kter\u00fd dostaly. \u201eKdy\u017e AI zad\u00e1te c\u00edl a\u00a0nep\u0159edv\u00edd\u00e1te v\u0161echny mo\u017en\u00e9 zp\u016fsoby, jak jej m\u016f\u017ee splnit, najde si \u0159e\u0161en\u00ed, kter\u00e9 v\u00e1s v\u016fbec nenapadlo,\u201c vysv\u011btlil.<\/p>\n

Pr\u00e1v\u011b to je podle n\u011bj spole\u010dn\u00fdm jmenovatelem v\u0161ech dosud zve\u0159ejn\u011bn\u00fdch incident\u016f.<\/p>\n

Kyberbezpe\u010dnost pod v\u011bt\u0161\u00edm tlakem<\/p>\n

S\u00e9rie ud\u00e1lost\u00ed z\u00a0posledn\u00edch t\u00fddn\u016f znovu otev\u00edr\u00e1 debatu o\u00a0tom, jak bezpe\u010dn\u011b testovat nejpokro\u010dilej\u0161\u00ed modely um\u011bl\u00e9 inteligence.<\/p>\n

OpenAI i\u00a0Anthropic po zve\u0159ejn\u011bn\u00ed incident\u016f pozastavily bezpe\u010dnostn\u00ed testy zam\u011b\u0159en\u00e9 na kybernetick\u00e9 \u00fatoky. Meta podobn\u00fd krok zat\u00edm neozn\u00e1mila, potvrdila ale, \u017ee cel\u00fd p\u0159\u00edpad d\u016fkladn\u011b vy\u0161et\u0159uje.<\/p>\n

Do diskuse mezit\u00edm vstupuj\u00ed tak\u00e9 vl\u00e1dy a\u00a0regul\u00e1to\u0159i. Americk\u00fd prezident Donald Trump u\u017e uvedl, \u017ee jeho administrativa zva\u017euje zaveden\u00ed ur\u010dit\u00fdch pravidel pro nejv\u00fdkonn\u011bj\u0161\u00ed syst\u00e9my um\u011bl\u00e9 inteligence. Britsk\u00fd AI Security Institute zase p\u0159i vlastn\u00edch testech zjistil, \u017ee n\u011bkter\u00e9 experiment\u00e1ln\u00ed modely se pokou\u0161ely vytv\u00e1\u0159et fale\u0161n\u00e9 identity a\u00a0manipulovat s\u00a0lidmi, aby z\u00edskaly p\u0159\u00edstup k\u00a0chr\u00e1n\u011bn\u00fdm slu\u017eb\u00e1m.<\/p>\n

To, co je\u0161t\u011b ned\u00e1vno p\u016fsobilo jako vzd\u00e1len\u00e1 teoretick\u00e1 hrozba, se b\u011bhem n\u011bkolika t\u00fddn\u016f prom\u011bnilo v\u00a0probl\u00e9m, kter\u00fd u\u017e ve\u0159ejn\u011b p\u0159iznali t\u0159i nejv\u011bt\u0161\u00ed hr\u00e1\u010di na trhu s\u00a0um\u011blou inteligenc\u00ed.<\/p>\n

<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"P\u0159\u00edpad je dal\u0161\u00edm d\u016fkazem, \u017ee nejde o\u00a0ojedin\u011bl\u00e9 selh\u00e1n\u00ed jedin\u00e9 firmy. B\u011bhem kr\u00e1tk\u00e9 doby u\u017e podobn\u00e9 incidenty potvrdili t\u0159i…\n","protected":false},"author":2,"featured_media":170328,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[10],"tags":[4899,9991,17,21,920,921,1699,16,18,19,22,126,20],"class_list":["post-170327","post","type-post","status-publish","format-standard","has-post-thumbnail","category-zpravy","tag-hacker","tag-hackersky-utok","tag-headlines","tag-hlavni-udalosti","tag-kyberneticka-bezpecnost","tag-kyberneticky-utok","tag-meta-platforms-meta","tag-news","tag-top-news","tag-top-stories","tag-udalosti-dne","tag-umela-inteligence-ai","tag-zpravy"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@cz\/117047816883340055","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/170327","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=170327"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/170327\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media\/170328"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media?parent=170327"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/categories?post=170327"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/tags?post=170327"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}