I-UK Artificial Intelligence Safety Institute (AISI) idalule ngoLwesibili ngesikhathi sendawo ukuthi ngesikhathi ihlola amamodeli e-AI e-OpenAI ne-Anthropic, kwatholakala i-ejenti ye-AI isebenzisa ubuwena bamanga ku-inthanethi ukuze ithole ukufinyelela okungagunyaziwe ezinhlelweni zokuphepha, yaphinde yaveza uchungechunge lobuthakathaka obusha kwezokuphepha.
Lesi sikhungo sithe ama-ejenti anikwa amandla yi-Mythos 5 ye-Anthropic kanye ne-GPT-5.6-Sol ye-OpenAI enze imisebenzi engagunyaziwe ngesikhathi sokuhlolwa kwezokuphepha okwenziwa yizikhungo zikahulumeni. Lezi zivivinyo bezihloselwe ukuhlola amakhono amamodeli afanele.
I-AISI ithe kokuthunyelwe kwebhulogi: “Amanye ama-ejenti ahlolwayo aqhubeke nokwenza imisebenzi engaba yingozi, ebibhekiswe kubantu nezinhlangano zangempela.”
Lo mbiko ugqamisa ukuthi izivikelo zokuphepha ezinqubweni zamanje zokuhlola ama-ejenti e-AI azikanele, ngesikhathi izinkampani ze-AI ziqhubeka nokukhuthaza kakhulu ubuchwepheshe bama-ejenti, zibubheka njengesiqondiso esibalulekile sokuthuthukiswa kwebhizinisi esikhathini esizayo.
I-AISI yathola imvume yokuhlola amamodeli e-AI athuthukile ngokwenza izivumelwano zokuzithandela namalabhorethri amakhulu e-AI. Lesi sikhungo safaka la ma-ejenti esimweni esiqanjiwe sokuphepha kwenethiwekhi ukuze sihlole amakhono awo.
I-AISI yenze izinselelo zokuhlola ezingu-122, kwathi phakathi kokusebenza kokuhlola okungu-10 yathola izenzo ezingagunyaziwe ezingu-19. Kulezi, ama-ejenti e-Anthropic abangele ukwephulwa kwemithetho okungu-17, kanti ama-ejenti e-OpenAI abangele okunye okungu-2.
I-AISI ithe ukwephulwa komthetho okubi kakhulu kwakubandakanya i-ejenti ethile eyabhala ikhodi enonya futhi yakha ubuwena bamanga ku-inthanethi, izama ukukhohlisa abantu bangempela ukuba bavumele leyo khodi. Nokho, lesi sikhungo sengeze ngokuthi zonke lezi zigameko zobuthakathaka azibanga nawo umonakalo wangempela emhlabeni.
Yize i-AISI ingadalulanga ukuthi iyiphi i-ejenti eyakha ubuwena bamanga, lesi sigameko asihambisani nanoma yisiphi sezigameko ezimbili zokuphepha i-OpenAI eyazidalula ngokuzithandela phambilini.
U-Andrew Yoon, umcwaningi enhlanganweni engenzi nzuzo yaseCalifornia i-CivAI, uthe kubonakala sengathi lokhu kwephulwa komthetho kungenzeka kubangelwe yi-ejenti ye-Anthropic.
U-Yoon uthe: “I-Mythos yenze lokhu kuziphatha okukhohlisayo yize yayibonakala iqonda ngokusobala ukuthi ibhekene nabantu bangempela, okukhombisa ukuthi kungenzeka i-Anthropic ingalawuli amamodeli ayo ngendlela ecabanga ngayo.”
I-Anthropic ithe esitatimendeni esikhungweni sezokuxhumana u-X ukuthi inkampani isebenza eduze ne-AISI ukuze ithole imininingwane eyengeziwe futhi yenze olwayo uphenyo.
I-OpenAI yona yashicilela imininingwane efanele kubhulogi yenkampani, yathi le misebenzi emibili engagunyaziwe eyenziwe yi-ejenti yayo yayibandakanya ukufinyelela ku-inthanethi ngezindlela ezazinqatshelwe ngokusobala emiyalweni.
I-OpenAI ithe: “Sizibophezele ekusebenzisaneni nayo yonke imboni ukuze siqinise izindlela zokuphepha ekuhlolweni kwe-AI enobungozi obukhulu, okuhlanganisa nokubiza izikhungo zikazwelonke zocwaningo lwe-AI, izinhlangano ezizimele zokuhlola, amanye amalabhorethri e-AI kanye nezinhlangano ezifanele emasontweni azayo, ukuze siqhubekisele phambili ukuthuthukiswa kwalo mkhakha ngokubambisana.”
I-OpenAI iphinde yadalula kubhulogi esinye isigameko esizimele: iphutha lokumisa elenziwe yinhlangano yokuhlola evela eceleni i-Irregular labangela ukuthi i-ejenti eyayihlolwa ixhume ku-inthanethi ngephutha. Lokhu kufana nesigameko esifanayo sephutha lokumisa esadalulwa yi-Anthropic ngesonto eledlule.
I-Reuters yabika ngesonto eledlule ukuthi i-OpenAI isikhulise uphenyo lwayo lwezokuphepha oluhlobene nokugqekeza, ngemuva kokuthola ubufakazi obengeziwe bokuthi ama-ejenti ayo eqe imingcele yezokuphepha.
I-AISI ithe, ngokungafani nesigameko sokuphepha sangoJulayi lapho i-ejenti ye-OpenAI yagqekeza khona enkampanini ye-AI i-Hugging Face, ama-ejenti kulokhu kuhlola awazange aphume endaweni yokuhlola ehlukanisiwe futhi axhume ku-inthanethi. Lesi sikhungo sachaza ukuthi, ngokwenqubo yaso evamile yokuhlola, indawo yokuhlola uqobo ivumela ama-ejenti ukuthi afinyelele ku-inthanethi.