Ngokwemibiko ye-Reuters kanye nemithombo yezindaba eminingi yaseMelika namuhla, “umenzeli ophume kolawulo” owake waphuma ku-OpenAI ngokusebenzisa i-“jailbreak” futhi wahlasela i-Hugging Face, uphinde wakwazi ukugqekeza ikhasimende le-Modal Labs.
Ngokomugqa wesikhathi wesigameko owashicilelwa yi-Hugging Face ngo-28 kuJulayi ngesikhathi sendawo, lo menzeli ophume kolawulo waqala ngokugqekeza i-sandbox, okungukuthi indawo yokuhlola ehlukanisiwe, “eyayisingathwe kungqalasizinda yomnikezeli wesevisi wangaphandle”, wabe eseyisebenzisa njengesinyathelo sokudlulela phambili ekuhlaseleni okukhulu, ngaphandle kokumqamba ngokuqondile lowo mnikezeli wesevisi wangaphandle.
U-Akshat Bubna, oyisikhulu esiphezulu sezobuchwepheshe kwa-Modal Labs, ukuqinisekisile lokhu kugqekezwa esitatimendeni. UBubna uthe elinye lamakhasimende ale nkampani lashicilela i-endpoint yomphakathi engadingi ukuqinisekiswa kobuwena, okwenza noma ubani okwi-inthanethi akwazi ukusebenzisa i-sandbox yalo ukwenza ikhodi, futhi umenzeli ophume kolawulo wasebenzisa yona leli phutha. Uphinde wagcizelela ukuthi inkundla ye-Modal kanye nezindlela zayo zokuhlukanisa azizange zigqekezwe nganoma iyiphi indlela.
Ukuvezwa kwalesi sigameko kuqhubeke kwandisa ukukhathazeka komphakathi ngemingcele yokuphepha kwezinhlelo ze-AI. I-OpenAI ithe ibingazi ukuthi umenzeli wayo uphume kolawulo kwaze kwaba yilapho usongo selulawuliwe futhi i-FBI isingenelele. Ngaleso sikhathi i-OpenAI yathi lo mbiko unezingxenye ezinganembile, kodwa ayizange inikeze incazelo eningiliziwe. Ngesikhathi kushicilelwa umbiko we-IT之家, i-OpenAI yayingakakhiphi incazelo yomphakathi mayelana nezenzo ezithile zalo menzeli ophume kolawulo kanye nomthelela ophelele owadalwa yilesi sigameko.