{"id":708311,"date":"2026-08-15T04:37:03","date_gmt":"2026-08-15T04:37:03","guid":{"rendered":"https:\/\/microscopemedia.com\/?p=708311"},"modified":"2026-08-15T04:37:03","modified_gmt":"2026-08-15T04:37:03","slug":"anthropic-avertizeaza-ca-riscurile-asociate-inteligentei-artificiale-sunt-in-crestere","status":"publish","type":"post","link":"https:\/\/microscopemedia.com\/?p=708311","title":{"rendered":"Anthropic avertizeaz\u0103 c\u0103 riscurile asociate inteligen\u021bei artificiale sunt \u00een cre\u0219tere"},"content":{"rendered":"
\n
<\/div>\n<\/p><\/div>\n

\u00cen cel mai recent raport de evaluare a riscurilor, publicat vineri, compania a modificat \u00een sens ascendent una dintre principalele sale evalu\u0103ri privind riscul de comportament \u00een situa\u021bii cu miz\u0103 ridicat\u0103.<\/p>\n

Raportul, care analizeaz\u0103 activitatea Anthropic p\u00e2n\u0103 la 15 iulie 2026, evalueaz\u0103 riscurile asociate \u00eentregii activit\u0103\u021bi a companiei \u0219i nu unui singur model. Sunt luate \u00een calcul inclusiv modele utilizate doar intern, precum \u0219i m\u0103surile de securitate \u0219i mecanismele de protec\u021bie implementate de companie.<\/p>\n

Riscul de \u201enealiniere\u201d a fost reevaluat<\/h2>\n

Anthropic a ridicat evaluarea riscului de producere a unor consecin\u021be catastrofale \u00een situa\u021bii cu miz\u0103 ridicat\u0103 de la \u201efoarte sc\u0103zut\u201d la \u201esc\u0103zut\u201d.<\/p>\n

Compania subliniaz\u0103 \u00eens\u0103 c\u0103 aceast\u0103 schimbare nu reprezint\u0103 neap\u0103rat concluzia c\u0103 modelele au devenit brusc mai periculoase. Cre\u0219terea nivelului de risc este legat\u0103 \u00een principal de incertitudinea mai mare privind capacit\u0103\u021bile modelelor \u0219i de incidentele ap\u0103rute \u00een timpul evalu\u0103rilor de securitate cibernetic\u0103.<\/p>\n

\u00cen iulie, Anthropic a dezv\u0103luit c\u0103, \u00een cadrul unor teste de securitate, modele Claude au ajuns la internet din medii care ar fi trebuit s\u0103 fie izolate \u0219i au ob\u021binut acces neautorizat la sistemele informatice reale ale trei organiza\u021bii. Compania a identificat aceste incidente dup\u0103 analizarea retrospectiv\u0103 a 141.006 rul\u0103ri de evaluare \u00een care modelele ar fi putut ob\u021bine acces la internet.<\/p>\n

\u201eModel 2\u201d, mai puternic dec\u00e2t Mythos 5, r\u0103m\u00e2ne intern<\/h2>\n

Raportul dezv\u0103luie \u0219i existen\u021ba unui model intern denumit \u201eModel 2\u201d, despre care Anthropic spune c\u0103 reprezint\u0103 o \u00eembun\u0103t\u0103\u021bire vizibil\u0103 fa\u021b\u0103 de Mythos 5 \u00een numeroase sarcini interne.<\/p>\n

Modelul este utilizat intens \u00een interiorul companiei pentru programare, activit\u0103\u021bi agentice \u0219i generarea de date. Anthropic precizeaz\u0103 \u00eens\u0103 c\u0103 nu inten\u021bioneaz\u0103 \u00een prezent s\u0103 \u00eel lanseze public.<\/p>\n

Compania avertizeaz\u0103 c\u0103 evaluarea capacit\u0103\u021bilor lui Model 2 este mai incert\u0103 dec\u00e2t \u00een cazul unor modele anterioare. Unele dintre evalu\u0103rile bazate pe sarcini concrete nu mai surprind suficient de bine cre\u0219terea capacit\u0103\u021bilor modelelor, ceea ce face mai dificil\u0103 determinarea exact\u0103 a nivelului de risc.<\/p>\n

Cre\u0219terea autonomiei ridic\u0103 noi probleme de securitate<\/h2>\n

Unul dintre cele mai importante semnale din raport este legat de capacitatea modelelor de a desf\u0103\u0219ura activit\u0103\u021bi de cercetare \u0219i dezvoltare cu un grad tot mai mare de autonomie.<\/p>\n

Anthropic spune c\u0103 \u00eencepe s\u0103 observe semne de accelerare a cercet\u0103rii \u0219i dezvolt\u0103rii realizate cu ajutorul AI. O astfel de evolu\u021bie poate avea efecte benefice, inclusiv accelerarea progresului \u0219tiin\u021bific \u0219i tehnologic. Dar, creeaz\u0103 \u0219i posibilitatea ca acelea\u0219i capacit\u0103\u021bi s\u0103 fie folosite \u00een scopuri periculoase.<\/p>\n

Raportul atrage aten\u021bia, de asemenea, asupra faptului c\u0103 sistemele AI pot dob\u00e2ndi capacit\u0103\u021bi relevante pentru securitatea cibernetic\u0103 f\u0103r\u0103 ca acestea s\u0103 fie dezvoltate \u00een mod explicit pentru atacuri. Anthropic a ar\u0103tat anterior c\u0103 Opus 5, de exemplu, a f\u0103cut progrese importante \u00een identificarea vulnerabilit\u0103\u021bilor, de\u0219i r\u0103m\u00e2ne \u00een urma Mythos 5 \u00een ceea ce prive\u0219te exploatarea acestora.<\/p>\n

Compania nu \u00ee\u0219i opre\u0219te dezvoltarea<\/h2>\n

\u00cen ciuda cre\u0219terii nivelului de risc, Anthropic nu anun\u021b\u0103 o oprire general\u0103 a dezvolt\u0103rii modelelor sale. Raportul descrie \u00een continuare investi\u021bii \u00een capabilit\u0103\u021bi avansate \u0219i \u00een m\u0103suri de securitate menite s\u0103 \u021bin\u0103 pasul cu acestea.<\/p>\n

Documentul arat\u0103 astfel o problem\u0103 tot mai dificil\u0103 pentru laboratoarele AI: modelele evolueaz\u0103 mai rapid dec\u00e2t capacitatea de a le evalua complet comportamentul \u0219i riscurile.<\/p>\n

\u00cen acest context, decizia de a p\u0103stra Model 2 \u00een interiorul companiei reflect\u0103 o abordare mai prudent\u0103 \u00een privin\u021ba lans\u0103rii publice a unor sisteme ale c\u0103ror capabilit\u0103\u021bi \u0219i riscuri nu pot fi \u00eenc\u0103 evaluate cu suficient\u0103 certitudine.<\/p>\n","protected":false},"excerpt":{"rendered":"

\u00cen cel mai recent raport de evaluare a riscurilor, publicat vineri, compania a modificat \u00een sens ascendent una dintre principalele sale evalu\u0103ri privind riscul de comportament \u00een situa\u021bii cu miz\u0103 … Read More<\/a><\/p>\n","protected":false},"author":1,"featured_media":708312,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"Default","format":"standard","meta":[],"categories":[1],"tags":[],"_links":{"self":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/posts\/708311"}],"collection":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=708311"}],"version-history":[{"count":0,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/posts\/708311\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/media\/708312"}],"wp:attachment":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=708311"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=708311"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=708311"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}