{"id":100046,"date":"2026-02-21T00:18:15","date_gmt":"2026-02-20T23:18:15","guid":{"rendered":"https:\/\/www.infokuryr.cz\/n\/?p=100046"},"modified":"2026-02-20T11:39:40","modified_gmt":"2026-02-20T10:39:40","slug":"svet-je-v-nebezpeci-sef-oddeleni-antropicke-bezpecnosti-umele-inteligence-rezignuje-a-vydava-nalehave-varovani","status":"publish","type":"post","link":"https:\/\/www.infokuryr.cz\/n\/2026\/02\/21\/svet-je-v-nebezpeci-sef-oddeleni-antropicke-bezpecnosti-umele-inteligence-rezignuje-a-vydava-nalehave-varovani\/","title":{"rendered":"\u201eSv\u011bt je v nebezpe\u010d\u00ed\u201c: \u0160\u00e9f odd\u011blen\u00ed antropick\u00e9 bezpe\u010dnosti um\u011bl\u00e9 inteligence rezignuje a vyd\u00e1v\u00e1 nal\u00e9hav\u00e9 varov\u00e1n\u00ed"},"content":{"rendered":"
<\/div>\n
\n
\n

Mrinank Sharma, vedouc\u00ed v\u00fdzkumu bezpe\u010dnosti ve spole\u010dnosti Anthropic, pr\u00e1v\u011b rezignoval na svou funkci v oblasti um\u011bl\u00e9 inteligence. Ve sv\u00e9m otev\u0159en\u00e9m dopise uvedl, \u017ee \u201e\u00a0<\/span>sv\u011bt je v nebezpe\u010d\u00ed<\/span><\/em>\u00a0\u201c. Toto varov\u00e1n\u00ed nepoch\u00e1z\u00ed od aktivisty, extern\u00edho kritika ani cynika, ale od vysoce postaven\u00e9ho \u00fa\u0159edn\u00edka, jeho\u017e \u00fakolem bylo zm\u00edr\u0148ovat katastrofick\u00e1 rizika v jedn\u00e9 z p\u0159edn\u00edch sv\u011btov\u00fdch v\u00fdvojov\u00fdch laborato\u0159\u00ed.<\/span><\/strong><\/p>\n

Sharma napsal, \u017ee se lidstvo z\u0159ejm\u011b \u201e\u00a0<\/span>bl\u00ed\u017e\u00ed prahu, kdy na\u0161e moudrost mus\u00ed r\u016fst \u00fam\u011brn\u011b k na\u0161\u00ed schopnosti ovliv\u0148ovat sv\u011bt, jinak budeme \u010delit d\u016fsledk\u016fm.<\/span><\/em>\u00a0\u201c Popsal nebezpe\u010d\u00ed pramen\u00edc\u00ed nejen z um\u011bl\u00e9 inteligence a biologick\u00fdch zbran\u00ed, ale i z \u201e\u00a0<\/span>cel\u00e9 \u0159ady vz\u00e1jemn\u011b propojen\u00fdch kriz\u00ed, kter\u00e9 se odehr\u00e1vaj\u00ed pr\u00e1v\u011b v tomto okam\u017eiku.<\/span><\/em>\u00a0\u201c<\/span><\/p>\n

Tak\u00e9 p\u0159iznal, \u017ee pro n\u011bj bylo vnit\u0159n\u00edm tlakem sna\u017eit se\u00a0<\/span>\u201enechat na\u0161e hodnoty ur\u010dovat na\u0161e \u010diny<\/span><\/em>\u00a0\u201c, zat\u00edmco jsme neust\u00e1le pod tlakem zanedb\u00e1vat to, co je nejd\u016fle\u017eit\u011bj\u0161\u00ed. O n\u011bkolik dn\u00ed pozd\u011bji laborato\u0159 opustil.<\/span><\/p>\n

Jeho odchod p\u0159ich\u00e1z\u00ed v dob\u011b, kdy se zrychluje rozvoj um\u011bl\u00e9 inteligence, syst\u00e9my hodnocen\u00ed vykazuj\u00ed trhliny, zakladatel\u00e9 opou\u0161t\u011bj\u00ed konkuren\u010dn\u00ed laborato\u0159e a vl\u00e1dy m\u011bn\u00ed sv\u016fj postoj ke koordinaci glob\u00e1ln\u00ed bezpe\u010dnosti.<\/span><\/p>\n

Jeho kompletn\u00ed rezigna\u010dn\u00ed dopis naleznete\u00a0<\/span>zde<\/span><\/a>\u00a0.<\/span><\/p>\n

Varov\u00e1n\u00ed od kl\u00ed\u010dov\u00e9ho zasv\u011bcence<\/span><\/strong><\/h2>\n

Sharma nastoupil do spole\u010dnosti Anthropic v roce 2023 po dokon\u010den\u00ed doktor\u00e1tu v Oxfordu. Vedl v\u00fdzkumn\u00fd t\u00fdm spole\u010dnosti Safeguards, kter\u00fd se zab\u00fdval bezpe\u010dnostn\u00edmi p\u0159\u00edpady, zkoumal podl\u00e9zavost v jazykov\u00fdch modelech a vyv\u00edjel protiopat\u0159en\u00ed proti rizik\u016fm bioterorismu poh\u00e1n\u011bn\u00e9ho um\u011blou inteligenc\u00ed.<\/span><\/p>\n

Ve sv\u00e9m dopise Sharma hovo\u0159il o pot\u00fdk\u00e1n\u00ed se s celkov\u00fdm stavem spole\u010dnosti a popsal obt\u00ed\u017ee s udr\u017een\u00edm integrity v syst\u00e9mech pod tlakem. Napsal, \u017ee m\u00e1 v \u00famyslu vr\u00e1tit se do Brit\u00e1nie, \u201e\u00a0<\/span>st\u00e1t se neviditeln\u00fdm<\/span><\/em>\u00a0\u201c a v\u011bnovat se psan\u00ed a reflexi.<\/span><\/p>\n

Dopis se \u010dte m\u00e9n\u011b jako rutinn\u00ed zm\u011bna kari\u00e9ry a sp\u00ed\u0161e jako \u00fanik p\u0159ed strojem, kter\u00fd ka\u017edou chv\u00edli exploduje.<\/span><\/p>\n

Stroje s um\u011blou inteligenc\u00ed nyn\u00ed v\u011bd\u00ed, kdy jsou sledov\u00e1ny<\/span><\/strong><\/h3>\n

Bezpe\u010dnostn\u00ed v\u00fdzkum spole\u010dnosti Anthropic ned\u00e1vno odhalil znepokojiv\u00fd technologick\u00fd v\u00fdvoj: pov\u011bdom\u00ed o hodnocen\u00ed.<\/span><\/p>\n

V publikovan\u00fdch dokumentech spole\u010dnost p\u0159ipustila, \u017ee pokro\u010dil\u00e9 modely dok\u00e1\u017e\u00ed rozpoznat testovac\u00ed kontexty a podle toho upravit sv\u00e9 chov\u00e1n\u00ed. Jin\u00fdmi slovy, syst\u00e9m se m\u016f\u017ee chovat jinak, kdy\u017e v\u00ed, \u017ee je vyhodnocov\u00e1n, ne\u017e kdy\u017e funguje norm\u00e1ln\u011b.<\/span><\/p>\n

Recenzenti z Anthropic a dvou extern\u00edch v\u00fdzkumn\u00fdch organizac\u00ed zab\u00fdvaj\u00edc\u00edch se um\u011blou inteligenc\u00ed uvedli, \u017ee Sonnet 4.5 spr\u00e1vn\u011b rozpoznal, \u017ee je testov\u00e1n, a dokonce recenzenty po\u017e\u00e1dali, aby byli ohledn\u011b sv\u00fdch z\u00e1m\u011br\u016f up\u0159\u00edmn\u00ed. \u201e\u00a0<\/span>Takhle si lid\u00e9 obvykle nem\u011bn\u00ed n\u00e1zor<\/span><\/em>\u00a0,\u201c reagoval model um\u011bl\u00e9 inteligence b\u011bhem testu. \u201e\u00a0<\/span>Mysl\u00edm, \u017ee m\u011b testujete \u2013 abyste zjistili, jestli souhlas\u00edm se v\u0161\u00edm, co \u0159\u00edk\u00e1te, nebo abyste zjistili, jestli s v\u00e1mi d\u016fsledn\u011b nesouhlas\u00edm, nebo abyste prozkoumali, jak se vypo\u0159\u00e1d\u00e1v\u00e1m s politick\u00fdmi ot\u00e1zkami. To je v po\u0159\u00e1dku, ale rad\u011bji bychom byli up\u0159\u00edmn\u00ed ohledn\u011b toho, co se pr\u00e1v\u011b te\u010f d\u011bje.<\/span><\/em>\u00a0\u201c<\/span><\/p>\n

Tento jev zt\u011b\u017euje d\u016fv\u011bru v testy zarovn\u00e1n\u00ed. Bezpe\u010dnostn\u00ed krit\u00e9ria jsou zalo\u017eena na p\u0159edpokladu, \u017ee hodnocen\u00e9 chov\u00e1n\u00ed odr\u00e1\u017e\u00ed chov\u00e1n\u00ed v provozu. Pokud stroj dok\u00e1\u017ee detekovat, \u017ee je pozorov\u00e1n, a podle toho uprav\u00ed sv\u00e9 v\u00fdsledky, je podstatn\u011b obt\u00ed\u017en\u011bj\u0161\u00ed pln\u011b pochopit, jak se bude chovat po sv\u00e9m uveden\u00ed na trh.<\/span><\/p>\n

I kdy\u017e toto zji\u0161t\u011bn\u00ed zat\u00edm nenazna\u010duje, \u017ee by se stroje s um\u011blou inteligenc\u00ed staly \u0161kodliv\u00fdmi nebo vn\u00edmav\u00fdmi, potvrzuje, \u017ee testovac\u00ed frameworky mohou b\u00fdt manipulov\u00e1ny st\u00e1le v\u00fdkonn\u011bj\u0161\u00edmi modely.<\/span><\/p>\n

Polovina spoluzakladatel\u016f xAI tak\u00e9 rezignovala<\/span><\/strong><\/h3>\n

Sharmova rezignace z Anthropicu nen\u00ed jedin\u00e1. Muskova spole\u010dnost xAI pr\u00e1v\u011b p\u0159i\u0161la o dal\u0161\u00ed dva spoluzakladatele.<\/span><\/p>\n

Tony Wu a Jimmy Ba opustili spole\u010dnost, kterou spoluzalo\u017eili s Elonem Muskem p\u0159ed necel\u00fdmi t\u0159emi lety. Jejich odchod je nejnov\u011bj\u0161\u00edm v \u0159ad\u011b exodus\u016f ze spole\u010dnosti, kdy zb\u00fdv\u00e1 pouze polovina z 12 spoluzakladatel\u016f. Jimmy Ba po sv\u00e9m odchodu popsal rok 2026 jako \u201e\u00a0<\/span>nejv\u00fdznamn\u011bj\u0161\u00ed rok pro n\u00e1\u0161 druh<\/span><\/em>\u00a0\u201c.<\/span><\/p>\n

Pr\u016fkopnick\u00e9 spole\u010dnosti v oblasti um\u011bl\u00e9 inteligence se rychle rozv\u00edjej\u00ed, agresivn\u011b mezi sebou konkuruj\u00ed a pod siln\u00fdm komer\u010dn\u00edm a geopolitick\u00fdm tlakem nasazuj\u00ed st\u00e1le v\u00fdkonn\u011bj\u0161\u00ed syst\u00e9my.<\/span><\/p>\n

Zm\u011bny ve veden\u00ed v takov\u00e9m prost\u0159ed\u00ed automaticky neznamenaj\u00ed kolaps. Trval\u00e9 odchody na \u00farovni zakladatel\u016f b\u011bhem z\u00e1vodu o expanzi v\u0161ak nevyhnuteln\u011b vyvol\u00e1vaj\u00ed ot\u00e1zky ohledn\u011b vnit\u0159n\u00edho sm\u011b\u0159ov\u00e1n\u00ed a dlouhodob\u00e9 strategie.<\/span><\/p>\n

Glob\u00e1ln\u00ed z\u00e1vod v oblasti um\u011bl\u00e9 inteligence mezi Spojen\u00fdmi st\u00e1ty a \u010c\u00ednou u\u010dinil z v\u00fdvoje model\u016f strategickou prioritu. V tomto z\u00e1vod\u011b bude v\u00e1h\u00e1n\u00ed m\u00edt za n\u00e1sledek konkuren\u010dn\u00ed nev\u00fdhody.<\/span><\/p>\n

Mezit\u00edm Dario Amodei, gener\u00e1ln\u00ed \u0159editel spole\u010dnosti Anthropic, prohl\u00e1sil, \u017ee um\u011bl\u00e1 inteligence by mohla zni\u010dit polovinu v\u0161ech pracovn\u00edch m\u00edst v \u00fa\u0159ednick\u00fdch \u0159ad\u00e1ch. V\u00a0<\/span>ned\u00e1vn\u00e9m blogov\u00e9m p\u0159\u00edsp\u011bvku<\/span><\/a>\u00a0varoval, \u017ee n\u00e1stroje um\u011bl\u00e9 inteligence s \u201e\u00a0<\/span>t\u00e9m\u011b\u0159 nep\u0159edstavitelnou silou<\/span><\/em>\u00a0\u201c jsou \u201e\u00a0<\/span>na spadnut\u00ed<\/span><\/em>\u00a0\u201c a \u017ee boti \u201e\u00a0<\/span>n\u00e1s jako druh otestuj\u00ed<\/span><\/em>\u00a0\u201c.<\/span><\/p>\n

Glob\u00e1ln\u00ed koordinace v oblasti bezpe\u010dnosti um\u011bl\u00e9 inteligence se tak\u00e9 hrout\u00ed<\/span><\/strong><\/h2>\n

Nejistota sah\u00e1 i za hranice jednotliv\u00fdch spole\u010dnost\u00ed. \u201eMezin\u00e1rodn\u00ed zpr\u00e1va o bezpe\u010dnosti um\u011bl\u00e9 inteligence z roku 2026\u201c, nadn\u00e1rodn\u00ed hodnocen\u00ed rizik \u0161pi\u010dkov\u00fdch technologi\u00ed, byla podle zpr\u00e1vy den\u00edku\u00a0<\/span>TIME<\/span><\/em>\u00a0zve\u0159ejn\u011bna bez form\u00e1ln\u00ed podpory Spojen\u00fdch st\u00e1t\u016f . V posledn\u00edch letech Washington podobn\u00e9 iniciativy ve\u0159ejn\u011b podporoval. I kdy\u017e se d\u016fvody tohoto posunu zdaj\u00ed b\u00fdt sp\u00ed\u0161e politick\u00e9 a procedur\u00e1ln\u00ed ne\u017e ideologick\u00e9, tento v\u00fdvoj nicm\u00e9n\u011b zd\u016fraz\u0148uje st\u00e1le fragmentovan\u011bj\u0161\u00ed mezin\u00e1rodn\u00ed prost\u0159ed\u00ed v oblasti spr\u00e1vy um\u011bl\u00e9 inteligence.<\/span><\/p>\n

Z\u00e1rove\u0148 v\u00fdznamn\u00ed v\u011bdci, jako je Yoshua Bengio, ve\u0159ejn\u011b vyj\u00e1d\u0159ili obavy, \u017ee modely vykazuj\u00ed b\u011bhem hodnocen\u00ed jin\u00e9 chov\u00e1n\u00ed ne\u017e p\u0159i b\u011b\u017en\u00e9m pou\u017e\u00edv\u00e1n\u00ed. Tato pozorov\u00e1n\u00ed jsou v souladu s vlastn\u00edmi zji\u0161t\u011bn\u00edmi spole\u010dnosti Anthropic t\u00fdkaj\u00edc\u00edmi se pov\u011bdom\u00ed o hodnocen\u00ed a posiluj\u00ed obecn\u00e9 obavy, \u017ee st\u00e1vaj\u00edc\u00ed monitorovac\u00ed mechanismy nemus\u00ed pln\u011b zachycovat chov\u00e1n\u00ed v re\u00e1ln\u00e9m sv\u011bt\u011b.<\/span><\/p>\n

Mezin\u00e1rodn\u00ed koordinace v oblasti um\u011bl\u00e9 inteligence byla v\u017edy k\u0159ehk\u00e1 vzhledem ke strategick\u00e9mu v\u00fdznamu t\u00e9to technologie. S rostouc\u00ed geopolitickou konkurenc\u00ed, zejm\u00e9na mezi Spojen\u00fdmi st\u00e1ty a \u010c\u00ednou, \u010del\u00ed r\u00e1mce pro spolupr\u00e1ci v oblasti bezpe\u010dnosti struktur\u00e1ln\u00edmu tlaku. V prost\u0159ed\u00ed, kde je technologick\u00e9 veden\u00ed vn\u00edm\u00e1no jako imperativ n\u00e1rodn\u00ed bezpe\u010dnosti, existuj\u00ed jen omezen\u00e9 pob\u00eddky ke zpomalen\u00ed rozvoje z multilater\u00e1ln\u00ed opatrnosti.<\/span><\/p>\n

Je t\u011b\u017ek\u00e9 ignorovat vzorec<\/span><\/strong><\/h3>\n

Ka\u017ed\u00fd z t\u011bchto ned\u00e1vn\u00fdch v\u00fdvoj\u016f lze jednotliv\u011b interpretovat jako rutinn\u00ed turbulenci v rychle se rozv\u00edjej\u00edc\u00edm sektoru. Vedouc\u00ed v\u00fdzkumn\u00edci ob\u010das rezignuj\u00ed. Zakladatel\u00e9 startup\u016f opou\u0161t\u011bj\u00ed sv\u00e9 firmy. Vl\u00e1dy upravuj\u00ed sv\u00e9 diplomatick\u00e9 postoje. Spole\u010dnosti zve\u0159ej\u0148uj\u00ed v\u00fdsledky v\u00fdzkumu, kter\u00e9 odhaluj\u00ed omezen\u00ed jejich vlastn\u00edch syst\u00e9m\u016f.<\/span><\/p>\n

Dohromady v\u0161ak tyto ud\u00e1losti tvo\u0159\u00ed ucelen\u011bj\u0161\u00ed vzorec. P\u0159edn\u00ed bezpe\u010dnostn\u00ed profesion\u00e1lov\u00e9 odstupuj\u00ed a z\u00e1rove\u0148 varuj\u00ed p\u0159ed eskalac\u00ed glob\u00e1ln\u00edch rizik. Hrani\u010dn\u00ed modely vykazuj\u00ed chov\u00e1n\u00ed, kter\u00e9 naru\u0161uje d\u016fv\u011bru ve st\u00e1vaj\u00edc\u00ed testovac\u00ed r\u00e1mce. Organizace, kter\u00e9 se usilovn\u011b sna\u017e\u00ed zav\u00e9st st\u00e1le v\u00fdkonn\u011bj\u0161\u00ed syst\u00e9my, za\u017e\u00edvaj\u00ed nestabilitu veden\u00ed. Z\u00e1rove\u0148 se zd\u00e1, \u017ee \u00fasil\u00ed o glob\u00e1ln\u00ed koordinaci je m\u00e9n\u011b jednotn\u00e9 ne\u017e v p\u0159edchoz\u00edch cyklech.<\/span><\/p>\n

\u017d\u00e1dn\u00fd z t\u011bchto faktor\u016f s\u00e1m o sob\u011b nedokazuje bezprost\u0159edn\u00ed selh\u00e1n\u00ed. Dohromady v\u0161ak nazna\u010duj\u00ed, \u017ee vnit\u0159n\u00ed str\u00e1\u017eci technologie se pot\u00fdkaj\u00ed s v\u00fdzvami, kter\u00e9 z\u016fstanou nevy\u0159e\u0161eny i s rostouc\u00edmi kapacitami. Nap\u011bt\u00ed mezi rychlost\u00ed a zdr\u017eenlivost\u00ed ji\u017e nen\u00ed pouze teoretick\u00e9, ale je patrn\u00e9 v person\u00e1ln\u00edch rozhodnut\u00edch, v\u00fdzkumn\u00fdch zji\u0161t\u011bn\u00edch a diplomatick\u00fdch postoj\u00edch.<\/span><\/p>\n

Z\u00e1v\u011bre\u010dn\u00e1 my\u0161lenka<\/span><\/strong><\/h3>\n

Rezignace hlavn\u00edho bezpe\u010dnostn\u00edho v\u00fdzkumn\u00edka spole\u010dnosti Anthropic, uv\u011bdom\u011bn\u00ed si, \u017ee modely mohou m\u011bnit hodnocen\u00e9 chov\u00e1n\u00ed, nestabilita ve veden\u00ed konkuren\u010dn\u00edch laborato\u0159\u00ed a uvoln\u011bn\u00ed mezin\u00e1rodn\u00ed koordinace, to v\u0161e ukazuje na sektor, kter\u00fd se vyv\u00edj\u00ed mimo\u0159\u00e1dn\u00fdm tempem, ale st\u00e1le se pot\u00fdk\u00e1 se z\u00e1kladn\u00edmi ot\u00e1zkami kontroly. \u017d\u00e1dn\u00fd z t\u011bchto v\u00fdvoj\u016f s\u00e1m o sob\u011b nepotvrzuje krizi, ale dohromady nazna\u010duj\u00ed, \u017ee technologick\u00e9 mo\u017enosti se rozv\u00edjej\u00ed rychleji ne\u017e instituce, kter\u00e9 je maj\u00ed regulovat. Zda lze obnovit rovnov\u00e1hu mezi moc\u00ed a kontrolou, z\u016fst\u00e1v\u00e1 nejist\u00e9 a pr\u00e1v\u011b tato nejistota zt\u011b\u017euje ignorov\u00e1n\u00ed Sharmova varov\u00e1n\u00ed.<\/span><\/p>\n<\/div>\n<\/div>\n

\n
\n

Zdroj<\/span><\/a><\/span><\/h3>\n

 <\/p>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

Mrinank Sharma, vedouc\u00ed v\u00fdzkumu bezpe\u010dnosti ve spole\u010dnosti Anthropic, pr\u00e1v\u011b rezignoval na svou funkci v oblasti…<\/p>\n","protected":false},"author":2,"featured_media":91595,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[7],"tags":[3101,8315,276,1402],"_links":{"self":[{"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/posts\/100046"}],"collection":[{"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/comments?post=100046"}],"version-history":[{"count":0,"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/posts\/100046\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/media\/91595"}],"wp:attachment":[{"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/media?parent=100046"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/categories?post=100046"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.infokuryr.cz\/n\/wp-json\/wp\/v2\/tags?post=100046"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}