{"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\/sq\/blog\/news\/anthropic-predupredila-chto-eyo-budushhaya-model-claude-mythos-stanet-hakerskoj-vundervaflej","title":{"rendered":"Anthropic paralajm\u00ebroi se modeli i saj i ardhsh\u00ebm Claude Mythos do t\u00eb b\u00ebhet \"arm\u00eb p\u00ebr hacker-at\"","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

Anthropic, OpenAI dhe kompani t\u00eb tjera teknologjike po p\u00ebrgatisin modele AI q\u00eb mund t\u00eb intensifikojn\u00eb ndjesh\u00ebm rrezikun e sulmeve masive kibernetike ndaj sistemeve korporative, shtet\u00ebrore dhe lokale. N\u00eb qend\u00ebr t\u00eb v\u00ebmendjes \u00ebsht\u00eb modeli i ri AI, Claude Mythos, i kompanis\u00eb Anthropic. Sipas Axios, kompania tashm\u00eb po paralajm\u00ebron zyrtar\u00ebt e lart\u00eb amerikan\u00eb se shfaqja e tij n\u00eb vitin 2026 rrit ndjesh\u00ebm probabilitetin e sulmeve t\u00eb k\u00ebtij niveli.<\/p>\n

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

Mythos p\u00ebrshkruhet si nj\u00eb sistem q\u00eb i lejon agjent\u00ebt AI t\u00eb dep\u00ebrtojn\u00eb n\u00eb m\u00ebnyr\u00eb autonome, me sakt\u00ebsi t\u00eb lart\u00eb dhe kompleksitet, n\u00eb sistemet korporative, shtet\u00ebrore dhe lokale. Drejtori i p\u00ebrgjithsh\u00ebm i Axios, Jim VandeHei, tha n\u00eb newsletterin e tij p\u00ebr lider\u00ebt se nj\u00eb nga burimet q\u00eb \u00ebsht\u00eb i njohur me modelet e ardhshme AI pranon nj\u00eb sulm t\u00eb madh kibernetik tashm\u00eb n\u00eb vitin 2026. Nd\u00ebrkoh\u00eb, biznesi \u00ebsht\u00eb p\u00ebrmendur si nj\u00eb objektiv ve\u00e7an\u00ebrisht t\u00eb ndjesh\u00ebm.<\/p>\n

M\u00eb par\u00eb Fortune studioi<\/a> materialet e rrjedhura nga blogu i Anthropic p\u00ebr Claude Mythos. N\u00eb t\u00eb, modeli AI quhet sistem q\u00eb \"aktualisht e tejkalon ndjesh\u00ebm \u00e7do model tjet\u00ebr AI n\u00eb kapacitetet kibernetike\". Po ashtu, thuhet se Mythos \"parashikon nj\u00eb val\u00eb t\u00eb ardhshme t\u00eb modeleve AI q\u00eb do t\u00eb jen\u00eb n\u00eb gjendje t\u00eb shfryt\u00ebzojn\u00eb dob\u00ebsit\u00eb n\u00eb nj\u00eb m\u00ebnyr\u00eb q\u00eb do t\u00eb tejkaloj\u00eb p\u00ebrpjekjet e mbrojtjes kibernetike\". Rreziku p\u00ebrkeq\u00ebsohet nga punonj\u00ebsit q\u00eb p\u00ebrdorin agjent\u00ebt AI pa kuptuar se kjo mund t\u00eb leht\u00ebsoj\u00eb qasjen e kriminel\u00ebve kibernetik\u00eb n\u00eb sistemet e brendshme t\u00eb kompanive t\u00eb tyre.<\/p>\n

N\u00eb fund t\u00eb vitit 2025, Anthropic revealed<\/a> do t\u00eb jet\u00eb rasti i par\u00eb i dokumentuar i nj\u00eb sulmi kibernetik, n\u00eb t\u00eb cilin pjes\u00ebrisht \u00ebsht\u00eb p\u00ebrdorur AI. Nj\u00eb grup i mb\u00ebshtetur nga shteti kinez p\u00ebrdori agjent\u00ebt AI p\u00ebr t\u00eb kryer autonomisht rreth 30 q\u00ebllime n\u00eb t\u00eb gjith\u00eb bot\u00ebn. AI zhvilloi vet\u00eb nga 80% n\u00eb 90% t\u00eb operacioneve taktike. Kjo ndodhi p\u00ebrpara se t\u00eb p\u00ebrmir\u00ebsoheshin ndjesh\u00ebm sistemet agjente dhe para shfaqjes s\u00eb m\u00ebnyrave t\u00eb reja dhe t\u00eb rrezikshme p\u00ebr p\u00ebrdorimin e tyre.<\/p>\n

Modelet e reja t\u00eb AI kan\u00eb nj\u00eb aft\u00ebsi m\u00eb t\u00eb lart\u00eb p\u00ebr t\u00eb mb\u00ebshtetur operimin autonom t\u00eb agjent\u00ebve AI. B\u00ebhet fjal\u00eb p\u00ebr sisteme q\u00eb lejojn\u00eb agjent\u00ebt t\u00eb mendojn\u00eb, veprojn\u00eb, arsyetojn\u00eb dhe improvizojn\u00eb n\u00eb m\u00ebnyr\u00eb t\u00eb pavarur pa pushime dhe lodhje. Shkalla e sulmeve tani p\u00ebrcaktohet kryesisht nga burimet kompjuterike t\u00eb agresor\u00ebve, jo nga numri i grupit t\u00eb tyre. Tani nj\u00eb person ka mund\u00ebsia<\/a> mund\u00ebsin\u00eb p\u00ebr t\u00eb realizuar sulme kibernetike, p\u00ebr t\u00eb cilat k\u00ebrkoheshin m\u00eb par\u00eb grupe t\u00eb t\u00ebra.<\/p>\n

Po nj\u00ebher\u00ebsh, po rritet edhe vulnerabiliteti i kompanive. Punonj\u00ebsit po p\u00ebrdorin gjithnj\u00eb e m\u00eb shum\u00eb Claude, Copilot dhe modele t\u00eb tjera agjent\u00ebsh, shpesh nga sht\u00ebpia, dhe krijojn\u00eb agjent\u00eb t\u00eb vet. N\u00eb shum\u00eb raste, k\u00ebta agjent\u00eb pa kontroll t\u00eb duhur fitojn\u00eb akses n\u00eb sistemet e brendshme t\u00eb pun\u00ebs. Kjo krijon nj\u00eb kanal tjet\u00ebr p\u00ebr kriminel\u00ebt kibernetik\u00eb p\u00ebr t\u00eb dep\u00ebrtuar. N\u00eb industri, kjo p\u00ebrdorim i paautorizuar ose i pakontrolluar i inteligjenc\u00ebs artificiale quhet \"IA n\u00eb hije\".<\/p>\n

Nj\u00eb anket\u00eb e burimit Dark Reading ka treguar se 48% e specialist\u00ebve t\u00eb siguris\u00eb kibernetike e konsiderojn\u00eb IA-n\u00eb agjente si vektorin kryesor t\u00eb sulmeve n\u00eb vitin 2026. Kjo k\u00ebrc\u00ebnim \u00ebsht\u00eb renditur m\u00eb lart se deepfake-t dhe mbi t\u00eb gjitha rreziqet e tjera t\u00eb p\u00ebrfshira n\u00eb anket\u00eb. \u00c7do punonj\u00ebs i \u00e7do kompanie n\u00eb bot\u00eb tashm\u00eb duhet t\u00eb kuptoj\u00eb rrezikun e p\u00ebrdorimit t\u00eb agjent\u00ebve t\u00eb IA pran\u00eb informacionit t\u00eb ndjesh\u00ebm. Ekipi teknik i Axios e sheh k\u00ebt\u00eb k\u00ebrc\u00ebnim si m\u00eb t\u00eb madhin p\u00ebr kompanin\u00eb n\u00eb k\u00ebt\u00eb moment. Konkluzioni praktik \u00ebsht\u00eb: p\u00ebr t\u00eb punuar me agjent\u00eb t\u00eb IA, nevojitet nj\u00eb mjedis i sigurt dhe i izoluar.<\/p>\n

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