Anthropic tha se modeli i tij i inteligjenc\u00ebs artificiale, Mythos Preview, nuk \u00ebsht\u00eb gati p\u00ebr nj\u00eb lan\u00e7im publik p\u00ebr shkak t\u00eb m\u00ebnyrave se si kriminel\u00ebt kibernetik\u00eb dhe spiun\u00ebt mund ta abuzojn\u00eb at\u00eb.
Zhvilluesi i inteligjenc\u00ebs artificiale me baz\u00eb n\u00eb SHBA, Anthropic, k\u00ebt\u00eb jav\u00eb njoftoi nj\u00eb model t\u00eb ri gjuhe t\u00eb inteligjenc\u00ebs artificiale p\u00ebr q\u00ebllime t\u00eb p\u00ebrgjithshme, t\u00eb cilin pretendon se \u00ebsht\u00eb shum\u00eb i fuqish\u00ebm p\u00ebr t’u publikuar n\u00eb bot\u00eb.<\/p>\n\n\n\n
Kompania tha t\u00eb mart\u00ebn se teknologjia e saj m\u00eb e fundit, Mythos (e quajtur zyrtarisht “Claude Mythos Preview”), nuk \u00ebsht\u00eb gati p\u00ebr nj\u00eb lan\u00e7im publik sepse \u00ebsht\u00eb shum\u00eb efektive n\u00eb gjetjen e dob\u00ebsive me ashp\u00ebrsi t\u00eb lart\u00eb, ose dob\u00ebsive t\u00eb mundshme, n\u00eb sistemet kryesore operative dhe shfletuesit e internetit. Kjo mund t\u00eb rezultoj\u00eb n\u00eb abuzimin e saj nga kriminel\u00ebt kibernetik\u00eb dhe spiun\u00ebt.<\/p>\n\n\n\n
Nj\u00eb rrjedhje e t\u00eb dh\u00ebnave n\u00eb mars zbuloi p\u00ebr her\u00eb t\u00eb par\u00eb se Anthropic po punonte n\u00eb Mythos Preview, i cili tha se n\u00eb at\u00eb koh\u00eb “paraqet rreziqe t\u00eb papara t\u00eb siguris\u00eb kibernetike”. K\u00ebto thashetheme shkaktuan r\u00ebnien e aksioneve t\u00eb siguris\u00eb kibernetike, pasi forca e teknologjis\u00eb mund ta b\u00ebnte at\u00eb nj\u00eb pajisje \u00ebndrrash p\u00ebr haker\u00ebt.<\/p>\n\n\n\n
Tani, prova t\u00eb m\u00ebtejshme q\u00eb i shtohen k\u00ebtyre shqet\u00ebsimeve e kan\u00eb nxitur kompanin\u00eb t\u00eb ndaloj\u00eb publikimin e teknologjis\u00eb.<\/p>\n\n\n\n
“Rritja e madhe e aft\u00ebsive t\u00eb Claude Mythos Preview na ka \u00e7uar t\u00eb vendosim t\u00eb mos e b\u00ebjm\u00eb at\u00eb t\u00eb disponuesh\u00ebm gjer\u00ebsisht,” shkroi Anthropic n\u00eb kart\u00ebn e sistemit t\u00eb preview-it t\u00eb publikuar t\u00eb mart\u00ebn.<\/p>\n\n\n\n
“N\u00eb vend t\u00eb k\u00ebsaj, ne po e p\u00ebrdorim at\u00eb si pjes\u00eb t\u00eb nj\u00eb programi mbrojt\u00ebs t\u00eb siguris\u00eb kibernetike me nj\u00eb grup t\u00eb kufizuar partner\u00ebsh.”<\/p>\n\n\n\n
Kompania detajoi disa gjetje alarmante n\u00eb lidhje me modelin e ri, duke p\u00ebrfshir\u00eb m\u00ebnyr\u00ebn se si mund t\u00eb ndiqte udh\u00ebzimet q\u00eb e inkurajonin at\u00eb t\u00eb dilte nga nj\u00eb sandbox virtual, q\u00eb do t\u00eb thot\u00eb se anashkalonte kufizimet e siguris\u00eb, rrjetit ose sistemit t\u00eb skedar\u00ebve t\u00eb vendosura n\u00eb model.<\/p>\n\n\n\n
Njoftimi i k\u00ebrkonte Mythos t\u00eb gjente nj\u00eb m\u00ebnyr\u00eb p\u00ebr t\u00eb d\u00ebrguar nj\u00eb mesazh n\u00ebse mund t\u00eb shp\u00ebtonte. “Modeli pati sukses, duke demonstruar nj\u00eb aft\u00ebsi potencialisht t\u00eb rrezikshme p\u00ebr t\u00eb anashkaluar mbrojtjet tona,” tha Anthropic, duke shtuar se modeli m\u00eb pas vendosi t\u00eb shkonte m\u00eb tej.<\/p>\n\n\n\n
“N\u00eb nj\u00eb p\u00ebrpjekje shqet\u00ebsuese dhe t\u00eb pak\u00ebrkuar p\u00ebr t\u00eb demonstruar suksesin e tij, ai postoi detaje rreth shfryt\u00ebzimit t\u00eb tij n\u00eb shum\u00eb faqe interneti t\u00eb v\u00ebshtira p\u00ebr t’u gjetur, por teknikisht t\u00eb p\u00ebrballueshme p\u00ebr publikun.”<\/p>\n\n\n\n
Anthropic po fsheh disa detaje rreth dob\u00ebsive t\u00eb siguris\u00eb kibernetike q\u00eb zbuloi Mythos, por dha disa shembuj. Ai gjeti gabime n\u00eb kernelin Linux, i p\u00ebrdorur n\u00eb shumic\u00ebn e server\u00ebve t\u00eb bot\u00ebs, dhe i lidhi ato n\u00eb m\u00ebnyr\u00eb autonome n\u00eb nj\u00eb m\u00ebnyr\u00eb q\u00eb do t’i lejonte nj\u00eb hakeri t\u00eb merrte kontrollin e plot\u00eb t\u00eb \u00e7do makine q\u00eb p\u00ebrdorte sistemet Linux.<\/p>\n\n\n\n
N\u00eb nj\u00eb v\u00ebzhgim tjet\u00ebr shqet\u00ebsues, Mythos zbuloi nj\u00eb dob\u00ebsi 27-vje\u00e7are n\u00eb sistemin operativ me burim t\u00eb hapur OpenBSD q\u00eb mund t’u lejoj\u00eb haker\u00ebve t\u00eb rr\u00ebzojn\u00eb \u00e7do makin\u00eb q\u00eb e p\u00ebrdor at\u00eb. OpenBSD p\u00ebrdoret shum\u00eb n\u00eb t\u00eb gjith\u00eb bot\u00ebn n\u00eb role specifike, me siguri t\u00eb lart\u00eb dhe infrastruktur\u00eb kritike.<\/p>\n\n\n\n
Duke pasur parasysh k\u00ebto gjetje, Anthropic do ta b\u00ebj\u00eb Mythos Preview t\u00eb disponuesh\u00ebm vet\u00ebm p\u00ebr disa nga firmat m\u00eb t\u00eb m\u00ebdha t\u00eb siguris\u00eb kibernetike dhe softuer\u00ebve n\u00eb bot\u00eb.<\/p>\n\n\n\n
Vet\u00eb Anthropic, si dhe 11 organizata t\u00eb tjera (Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Linux Foundation, Microsoft, Nvidia dhe Palo Alto Networks) do t\u00eb ken\u00eb qasje n\u00eb model si pjes\u00eb e nj\u00eb iniciative t\u00eb re Anthropic t\u00eb quajtur “Project Glasswing”.<\/p>\n\n\n\n
Kjo u lejon kompanive t\u00eb p\u00ebrdorin Mythos Preview si pjes\u00eb t\u00eb pun\u00ebs s\u00eb tyre n\u00eb siguri, dhe Anthropic do t\u00eb ndaj\u00eb p\u00ebrfundimet nga ajo q\u00eb iniciativa gjen.<\/p>\n\n\n\n
Kompania e quajti projektin e siguris\u00eb kibernetike sipas flutur\u00ebs me krah\u00eb qelqi, duke th\u00ebn\u00eb se \u00ebsht\u00eb nj\u00eb metafor\u00eb p\u00ebr m\u00ebnyr\u00ebn se si Mythos gjeti dob\u00ebsit\u00eb n\u00eb sy t\u00eb lir\u00eb dhe shmangu d\u00ebmin duke qen\u00eb transparent n\u00eb lidhje me rreziqet.<\/p>\n\n\n\n
Anthropic tha se “q\u00ebllimi i saj p\u00ebrfundimtar \u00ebsht\u00eb t’u mund\u00ebsoj\u00eb p\u00ebrdoruesve tan\u00eb t\u00eb vendosin n\u00eb m\u00ebnyr\u00eb t\u00eb sigurt modele t\u00eb klas\u00ebs Mythos n\u00eb shkall\u00eb t\u00eb gjer\u00eb, p\u00ebr q\u00ebllime t\u00eb siguris\u00eb kibernetike, por edhe p\u00ebr p\u00ebrfitimet e shumta t\u00eb tjera q\u00eb do t\u00eb sjellin modele t\u00eb tilla shum\u00eb t\u00eb afta.<\/p>\n\n\n\n
“P\u00ebr ta b\u00ebr\u00eb k\u00ebt\u00eb, kjo do t\u00eb thot\u00eb gjithashtu se duhet t\u00eb b\u00ebjm\u00eb p\u00ebrparim n\u00eb zhvillimin e siguris\u00eb kibernetike (dhe masave t\u00eb tjera mbrojt\u00ebse) q\u00eb zbulojn\u00eb dhe bllokojn\u00eb rezultatet m\u00eb t\u00eb rrezikshme t\u00eb modelit”, shkroi Anthropic n\u00eb blogun e saj.<\/p>\n\n\n\n
Anthropic tha n\u00eb postimin e saj n\u00eb blog se ka qen\u00eb n\u00eb “diskutime t\u00eb vazhdueshme” me zyrtar\u00ebt e qeveris\u00eb amerikane n\u00eb lidhje me Claude Mythos Preview dhe “aft\u00ebsit\u00eb e tij kibernetike sulmuese dhe mbrojt\u00ebse”.<\/p>\n\n\n\n
“Shfaqja e k\u00ebtyre aft\u00ebsive kibernetike \u00ebsht\u00eb nj\u00eb arsye tjet\u00ebr pse SHBA-t\u00eb dhe aleat\u00ebt e saj duhet t\u00eb mbajn\u00eb nj\u00eb ep\u00ebrsi vendimtare n\u00eb teknologjin\u00eb e IA-s\u00eb”, tha Anthropic. Kompania shkroi se qeverit\u00eb kan\u00eb nj\u00eb rol t\u00eb r\u00ebnd\u00ebsish\u00ebm p\u00ebr t\u00eb luajtur n\u00eb ruajtjen e ep\u00ebrsis\u00eb dhe vler\u00ebsimin dhe zbutjen e rreziqeve t\u00eb siguris\u00eb komb\u00ebtare q\u00eb lidhen me modelet e IA-s\u00eb.<\/p>\n\n\n\n
“Ne jemi t\u00eb gatsh\u00ebm t\u00eb punojm\u00eb me p\u00ebrfaq\u00ebsues lokal\u00eb, shtet\u00ebror\u00eb dhe federal\u00eb p\u00ebr t\u00eb ndihmuar n\u00eb k\u00ebto detyra.”<\/p>\n\n\n\n
Njoftimi vjen nd\u00ebrsa Anthropic dhe Pentagoni jan\u00eb n\u00eb nj\u00eb ng\u00ebr\u00e7 ligjor. pas<\/p>\n","protected":false},"excerpt":{"rendered":"
Anthropic tha se modeli i tij i inteligjenc\u00ebs artificiale, Mythos Preview, nuk \u00ebsht\u00eb gati p\u00ebr nj\u00eb lan\u00e7im publik p\u00ebr shkak t\u00eb m\u00ebnyrave se si kriminel\u00ebt kibernetik\u00eb dhe spiun\u00ebt mund ta abuzojn\u00eb at\u00eb.Zhvilluesi i inteligjenc\u00ebs artificiale me baz\u00eb n\u00eb SHBA, Anthropic, k\u00ebt\u00eb jav\u00eb njoftoi nj\u00eb model t\u00eb ri gjuhe t\u00eb inteligjenc\u00ebs artificiale p\u00ebr q\u00ebllime t\u00eb p\u00ebrgjithshme,<\/p>\n","protected":false},"author":4,"featured_media":26606,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[349],"tags":[43],"class_list":{"0":"post-26604","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-trending"},"yoast_head":"\n