- Abacwaningi baseCalif basebenzise i-Claude Mythos Preview ka-Anthropic ukwakha i-exploit esebenzayo ngokumelene ne-M5 Memory Integrity Enforcement ye-Apple esikhathini esingaphansi kwesonto.
- Lokhu kuhlasela kuhlanganisa amaphutha amabili kanye namasu athuthukile okubangela ukonakala kwememori ye-macOS kernel nokufinyelela ezindaweni zesistimu ezinqunyelwe.
- Ochwepheshe bachaza lokhu njengokusebenzisa umphakathi kokuqala okusinda ngempumelelo ekuvikelweni kwe-MIE okusha kwe-Apple okusizwa yihadiwe kuma-chip e-M5.
- Leli cala libhebhethekisa impikiswano ngokufinyelela okulawulwayo kumamodeli e-AI anamandla angaqinisa futhi anciphise ukuphepha kwe-inthanethi.
Ezinsukwini zamuva nje, ukuhlolwa kokuphepha kwe-inthanethi okungabonakali kahle sekuphenduke iphuzu elikhulu lokuxoxa emphakathini wezokuphepha: inguqulo yokubuka kuqala yemodeli ye-AI ethuthukisiwe ye-Anthropic, I-Claude Mythos, idunyiswe ngokusiza ukuphula esinye sezivikelo zehadiwe ezintsha ze-Apple kuma-chip ayo e-M5. Lesi siqephu sibhekwa njengesibonelo esicacile sendlela i-AI yanamuhla engaveza ngayo amaphutha abalulekile futhi ingase ishintshe ibhalansi phakathi kwabahlaseli nabavikeli.
Ngokusho kwabacwaningi enkampanini entsha esePalo Alto, iCalif, Imodeli yokubuka kuqala ye-Mythos idlale indima ebalulekile ekusekeleni ekwakheni isenzo esisebenzayo esibeka engcupheni ukuvikelwa kwe-Memory Integrity Enforcement (MIE) yakamuva ye-Apple kuma-Mac asekelwe ku-M5. Nakuba lokhu kuhlasela kusadinga ulwazi olujulile lwabantu, ithimba lithi i-AI isheshise kakhulu ukutholakala kwamaphutha angasetshenziswa, iphakamisa imibuzo emisha mayelana nokuthi ubani okufanele avunyelwe ukusebenzisa lezo zinhlelo futhi ngaphansi kwaziphi izimo.
Indlela uClaude Mythos asize ngayo ukwephula ukuvikelwa kwe-M5 kwe-Apple
Ingqikithi yendaba igxile eCalif, inkampani yezokuphepha kwe-inthanethi esesikhungweni sezobuchwepheshe saseCalifornia iPalo Alto, eyayihlose ukuhlola ukuthi i-Apple iphumelela kangakanani. indlela entsha ye-MIE kuma-chip e-M5 ingakwazi ukumelana namasu okuxhaphaza esimanje. Phakathi nalolo cwaningo, ithimba lisebenzise i-Claude Mythos Preview ka-Anthropic njengohlobo lomshayeli we-AI osizayo ukuhlaziya izindlela zekhodi, ukucabanga ngobuthakathaka obungaba khona, kanye nokubeka phambili izindawo zokuhlasela ezithembisayo.
Abacwaningi baseCalif babika ukuthi ekugcineni bakwazile ukuhlanganisa amaphutha amabili ahlukene esofthiwe anamasu amaningana akhethekile okugenca ukuze kufezwe ukonakala kwememori kwezinga le-kernel ku-macOS esebenza kwihadiwe ye-M5. Lokhu konakala kwabavumela ukuthi bafinyelele izingxenye zesistimu ezivame ukuvinjelwa, bedlula ngempumelelo ezinye zezivikelo ezithuthukisiwe kakhulu ezisizwa yihadiwe ye-Apple.
Okwenza leli cala livelele ukuthi ukuxhashazwa okubangelwa yilokho kubonakala sengathi, ngamazwi acashunwe yimibiko eminingi, ukuhlaselwa kokuqala kwenkohlakalo yememori ye-macOS kernel okwaziwa esidlangalaleni lokho kusinda ekuvikelweni kwe-MIE kwe-Apple kuma-chip e-M5. Ngenxa yendlela yokuzivikela i-Apple ebilokhu iyiphakamisa njengentuthuko enkulu ngemuva kweminyaka yokutshalwa kwezimali, ukubona ukuxhashazwa komphakathi kwenzeka ngokushesha kangaka kuye kwaheha ukunaka kwemboni.
Imithombo ehlanganisa i-Wall Street Journal kanye nezitolo ezigxile ku-Mac zigcizelele ukuthi lokhu kuxhashazwa akuncikile ekuhlulekeni kwehadiwe okungavamile noma amaqhinga okufinyelela ngokomzimba. Esikhundleni salokho, kuhlanganisa amaphutha anengqondo, uma esetholakele, angaguqulwa abe yindlela yokuhlasela eqhubekayo ekwazi ukunciphisa ukuvikelwa kwe-MIE ezinhlelweni ezithintekile.
I-Calif ngokwayo ichaze lo msebenzi njengokuhlolwa kwalokho okwenzekayo lapho Amathuluzi e-AI aphezulu ahambisana nabathuthukisi abanolwazi lokuxhaphaza, kunokuba kube yisimo sokugenca okuzenzakalelayo ngokuchofoza okukodwa. Noma kunjalo, isivinini kanye nobuchwepheshe bomphumela kuphakamise izexwayiso mayelana nokuthi amakhono anjalo angasakazeka ngokushesha kangakanani uma engalawulwa ngokuqinile.
Kuyini ukuqiniswa kobuqotho bememori ku-chip ye-M5?
Kwethulwe i-Apple Memory Integrity Enforcement njengesici sokuphepha esiyinhloko somugqa wayo we-chip we-M5, esenzelwe ukuqinisa amadivayisi ngokumelene nezigaba zonke zokuhlaselwa kokonakala kwememori. Umqondo ulula kodwa unesifiso esikhulu: sebenzisa ukwesekwa kwehadiwe okuzinikele ukuqapha nokulawula indlela imemori esetshenziswa ngayo, okwenza kube nzima kakhulu ngabahlaseli ukuduna ukwenziwa kwekhodi esezingeni eliphansi.
Ngokusho kokuhlaziywa kukaCalif, i-Apple ichithe cishe iminyaka emihlanu iletha i-MIE kusukela kumkhiqizo womkhiqizo kuya kowokuthumela, ngokutshalwa kwezimali okungenzeka ukuthi kufinyelele ezigidigidini zamaRandi. Le nkampani ibeke i-MIE njengoshintsho lwesinyathelo se- ukuphepha kwememori okusizwa yihadiwe, ihlose ukususa amasu okuxhaphaza asenesikhathi eside ayesetshenziswa ezizukulwaneni zangaphambilini zamadivayisi e-iOS kanye ne-macOS.
Ngaphambi kocwaningo lukaCalif, i-MIE ibike ukuthi iphumelele kakhulu ekusebenzeni. Le nkampani entsha ithi yonke iketanga lokuxhaphaza elaziwa esidlangalaleni Ukuqondisa izinhlelo ze-iOS zamanje kuphazanyiswe yi-MIE, okuhlanganisa nama-kit adumile njengephakheji ye-Coruna ne-Darksword esanda kuvuza. Ngamanye amazwi, i-MIE ibhekwe njengesivimbelo esikhulu sabahlaseli abathembele kumaphethini okonakala kwememori yakudala.
Ngokuphikisana nalokhu, ukuvela kokuxhashazwa kwenkumbulo yomphakathi ye-kernel okungasebenza naphezu kokuvikelwa kwe-MIE kuhadiwe ye-M5 kubhekwa njengentuthuko ebalulekile yobuchwepheshe nengokomfanekiso. Akusho ukuthi i-MIE yehlulekile iyonke, kodwa kuyakhombisa ukuthi abacwaningi abazimisele basengakwazi ukuthola izindlela zokuhlasela ezisebenzayo, ikakhulukazi uma kusekelwa amathuluzi okuhlaziya anamandla njengoClaude Mythos.
Imininingwane yobuchwepheshe isagcinwe eduze; iCalif ne-Anthropic abakakhiphi imibhalo ephelele yochungechunge lokuxhaphaza, begcizelela ukuthi balinde kuze kube yilapho i-Apple isixazulule ngokuphelele lolu daba. Kodwa-ke, ukuphawula kwabo kusikisela ukuthi yize i-MIE iphakamisa izinga kakhulu, akuyona isivikelo esingenakuphulwa, futhi izindlela zokunciphisa izingozi zesikhathi esizayo kuzodingeka zicabange ukuthi Ukutholakala kobuthakathaka obusheshisiwe be-AI kuyingxenye yendawo eyingozi.
Umugqa wesikhathi: kusukela ekutholakaleni kwamaphutha kuya ekuxhashazweni okusebenzayo esikhathini esingaphansi kwesonto
Esinye sezici eziphawuleka kakhulu endabeni kaCalif ijubane lentuthuko. Le nkampani ibeke isikhathi esinembile esibonisa ukuthi konke lokhu kuxhashazwa—kusukela ekuhlonzweni kokuqala kwesiphazamisi kuya ebufakazini obusebenzayo obuthembekile—kwahlangana esikhathini esingaphansi kwezinsuku eziyisikhombisa, isivinini ngisho nochwepheshe bezokuphepha abaqinile abasithola siphawuleka.
Njengoba kuchaziwe yithimba, inqubo yaqala lapho umcwaningi uBruce Dang wabona amaphutha okuqala afanele ngo-Ephreli 25. Phakathi nezinsuku ezimbalwa, uchwepheshe wezokuphepha UDion Blazakis bajoyine lo mzamo ngo-Ephreli 27, beletha ubuchwepheshe obengeziwe bokubumba isu lokuhlasela elishintshayo nokuqonda kangcono ukuthi i-MIE yasabela kanjani emizamweni yokuqala yokuxhashazwa.
Ngemva nje kwalokho, unjiniyela UJosh Maine wagxila ekwakheni amathuluzi adingekayo ukuhlola, ukuhlola, kanye nokulungisa izindlela zokuxhaphaza eziphakanyiswe abacwaningi abangabantu kanye nemodeli ye-Mythos. Ngomhlaka-1 Meyi, ezinsukwini ezimbalwa nje ngemva kokutholwa kokuqala, i-Calif ithi yayine-kernel exploit esebenza ngokugcwele ekwazi ukudlula i-MIE ezinhlelweni ezisekelwe ku-M5 eziqondiwe.
Abacwaningi bagcizelela ukuthi lokhu kuguquka okusheshayo kwakungewona nje umphumela wokwenza izinto ngokuzenzakalela. Esikhundleni salokho, bachaza umsebenzi wokubuyela emuva nokubuyela emuva lapho Ukubuka Kuqala kwe-Mythos kuveze amaphethini athembisayo obuthakathaka, okwaqinisekiswa ochwepheshe babantu, bakulungisa, futhi bakubopha ngendlela ehambisanayo. Umjikelezo oqinile phakathi kwezinkolelo-mbono ezikhiqizwe yi-AI kanye nokuhlaziywa kochwepheshe kubonakala sengathi kucindezele lokho obekungathatha amasonto noma izinyanga kwaba yindaba yezinsuku ezimbalwa.
UCalif uchaze le mpumelelo njengokubonakaliswa kokuqala kodwa okusobala kokuthi “kwenzekani lapho ubuhlakani bomshini kanye nolwazi lwabantu kuhlangana” emkhakheni wezokuphepha ohlaselayo. Ukuthola i-kernel exploit ngokumelene nalokho abakubheka njengokuvikela okuqine kakhulu kwe-Apple “ngesonto”, ngamazwi abo, kuwuphawu lokuthi lokhu kuhlangana cishe kuzoba amandla amakhulu—okuzodingeka abavikeli babhekane nakho kusukela ekuqaleni.
Indima kaClaude Mythos: enamandla, kodwa ayizimele ngokuphelele
Okubalulekile kule ndaba yi-Anthropic's Ukubuka KukaClaude Mythos, imodeli engakakhishwa emphakathini ngenxa yamakhono ayo athuthukile ekuhlaziyweni nasekusetshenzisweni kwezinkinga zesofthiwe. Esikhundleni salokho, ukufinyelela kukhawulelwe eqenjini elivaliwe lezinkampani ezinkulu zobuchwepheshe, izikhungo zezezimali, kanye nezinhlangano zocwaningo ezihlolwe ngaphansi kohlelo lwe-Anthropic's Project Glasswing.
UCalif ugcizelele ngokuphindaphindiwe ukuthi uhlelo lwe-AI lunegalelo elikhulu kuzo zonke izigaba zentuthuko yalo msebenzi. Ngokusho kwenkampani entsha, i-Mythos ibonise ikhono lokubona ngokushesha lapho izimbungulu ezintsha zingena khona. amakilasi okuxhaphaza aqondakala kahle, kanye nokusikisela izindlela lapho lobo buthakathaka bungase buhlanganiswe ndawonye noma buphathwe kabi. La mandla okuqaphela amaphethini avumele ithimba ukuthi lidlule izinqwaba zamaphutha amaningi ngesandla.
Le nkampani ichaze i-Mythos ngolimi oluqokomisa amandla ayo kanye nokujwayelekile kwayo: uma imodeli isifundile ukuhlasela okuthile isigaba sezinkinga zokuphepha, kungenza lolo lwazi luvame ukufinyeleleka ezinhlosweni eziningi ezifanayo. Lokho kusho ukuthi amasu afanayo omqondo asebenza engxenyeni eyodwa yesofthiwe angadluliselwa, ngokulungiswa, ezinhlelweni ezihluke ngokuphelele.
Noma kunjalo, abacwaningi baseCalif bayaqikelela ukugcizelela lokho ukwahlulela komuntu kwahlala kuyisisekelo. I-MIE kuma-chip e-M5 iyindlela entsha yokunciphisa, futhi izingxenye zokuziphatha kwayo ngaphansi kokuhlaselwa okusebenzayo zazingakacaciswa ngokugcwele, ngisho nangaphakathi kwa-Apple. Ngenxa yalokho, ukuqonda ukuthi ungadlula kanjani noma usebenzele kanjani lezo zivikelo kudinga onjiniyela abanolwazi lokuxhaphaza ukuze bahumushe iziphakamiso ze-AI, bazihlole ngokumelene nehadiwe yangempela, futhi baphinde basebenzise impendulo ecashile evela ekuphahlazekeni, amalogi, kanye nokuziphatha kwesistimu.
Ithimba lizame ngamandla alo onke ukumelana nomqondo wokuthi iMythos “iwumkhiqizi wokuxhaphaza osebenzisa inkinobho yokucindezelwa.” Kunalokho, bayichaza njengomsizi onekhono elikhulu okhulisa amakhono abantu—ikakhulukazi ngokufinyeza isikhathi esidingekayo ukuze basuke ku- ukutholakala kwezinambuzane kuya ekusetshenzisweni okungokoqoboNoma kunjalo, iqiniso lokuthi ithuluzi elinjalo likhona, futhi lingasebenza kahle kangaka ezandleni zochwepheshe, livusa impikiswano ebanzi mayelana nokulawulwa kokufinyelela kanye nokusetshenziswa okunomthwalo wemfanelo.
Ukusabela kwemboni kanye nempendulo ye-Apple
Lokhu kutholwa kudonsele isithakazelo esikhulu emhlabeni wonke wezobuchwepheshe, kokubili ngenxa yobuchwepheshe balo msebenzi kanye nendlela owakhiwe ngayo. Iziteshi zabezindaba ezifana ne- The Wall Street Journal babeke leli cala njengoshintsho oluphawulekayo emandleni okuhlasela kwe-cybersecurity, lapho i-AI ishintsha isuka ekukhathazekeni okungokwemfundiso iye ekubangeleni ukuhlaselwa okuphezulu.
Ngasohlangothini lwe-Apple, ukuphawula komphakathi kuhlale kuqaphile. Esitatimendeni esicashunwe yi-WSJ, ummeleli we-Apple uthembele emgqeni ojwayelekile wenkampani wokuthi ukuphepha komsebenzisi kuyinto eza kuqala nokuthi noma yibuphi ubuthakathaka obubikiwe buphenywa kahle. Ngaphandle kwalokho, inkampani ayizange inikeze imibono eningiliziwe ngokuxhashazwa kweCalif noma ukuthi ingaguqula kanjani ukuphindaphinda kwe-MIE esikhathini esizayo.
Nokho, kunezinkomba zokuthi i-Apple isiqalile ukuthatha lokhu okutholakele ngokungathi sína. Ababukeli baphawule izinkomba kukho kokubili. Ucwaningo lwe-Calif kanye ne-Anthropic kumanothi okukhululwa kwe-macOS Tahoe 26.5, isibuyekezo esikhishwe ngoMsombuluko wamuva nje. Lokhu kuhunyushwe njengophawu lokuthi okungenani umsebenzi othile wokunciphisa uyaqhubeka, yize iCalif noma i-Apple bengakaqinisekisi ukuthi yiziphi izici zokuxhashazwa, uma zikhona, ezilungisiwe ngokuphelele.
ICalif, yona, igcizelele ukuthi ibambezela amadokhumenti obuchwepheshe aphelele kuze kube yilapho i-Apple “ekugcineni isiqede i-vector yokuhlasela.” Le nkampani ikhethe ukwenza kanjalo. ukudalula ubuthakathaka mathupha endlunkulu yaseCalifornia ye-Apple esikhundleni sokuthembela eziteshini ezijwayelekile zokuthumela amaphutha, becaphuna isifiso sokugwema ukulahleka enqwabeni yemibiko etholwa abathengisi abakhulu—ikakhulukazi mayelana nemicimbi efana ne-Pwn2Own, lapho ukudalulwa okuhleliwe kungaba yisiphithiphithi kalula.
Le ndlela yokudalula ubuso nobuso isho ukuthi zonke izinhlangothi ziyaqaphela ukuzwela kokuxhashazwa okungagcini nje ngokubhekisisa ukwakheka kwe-chip okusha kwe-Apple kodwa futhi okubonisa indima ekhulayo ye-AI ekudaluleni amaphutha anjalo. Kuze kube yilapho kuvela imininingwane eyengeziwe, iningi lomphakathi obanzi lisala lifunda phakathi kwemigqa yezitatimende zomphakathi ezimfushane kanye namacebo amancane obuchwepheshe.
Kungani i-Anthropic yayivimbela ukufinyelela kuClaude Mythos
Isinqumo sika-Anthropic sokugcina uClaude Mythos uphume ezandleni zomphakathi jikelele Kungaphambi kwesiqephu se-Calif-Apple, kodwa lesi sigameko sibhekwa njengokuqinisekiswa kwalezo zindlela zokuqapha. Ngesikhathi sokuhlolwa kwangaphakathi nokuhlolwa kwangaphandle, inkampani kubikwa ukuthi iphawule ukuthi i-Mythos ingazibona ngokuzimela futhi, kwezinye izimo, isize ekusebenziseni ubuthakathaka besofthiwe ezingeni elingaphezu kwanoma yiluphi uhlelo lwe-AI lomphakathi lwangaphambilini.
Ngenxa yalezo ziphumo, i-Anthropic ikhethe ukukhishwa okulawulwa ngokuqinile. Ngaphansi kwesinyathelo ngezinye izikhathi esibizwa ngokuthi Iphrojekthi ye-Glasswing, I-Mythos itholakala kuphela kubalingani abakhethiwe—izinkampani ezinkulu zobuchwepheshe, amabhange, kanye nohlu olufushane lwezinhlangano zocwaningo—abavumayo ukusebenza ngaphansi kweziqondiso eziqinile zokusebenzisa. Inhloso ukusebenzisa amandla ale modeli ekuvikelekeni nasekucwaningeni kokuzivikela, ngenkathi kunciphisa ingozi yokuthi ingasetshenziswa kabi ngabahlaseli abasebenzisa amathuba.
Leli cebo alikaqedi ngokuphelele ukukhathazeka. Abanye ababukeli bemboni bathi ngisho nokufinyelela okulinganiselwe kuthuluzi elinjalo kungasheshisa ukuguquguquka "komncintiswano wezikhali", lapho izinhlangano ezinokufinyelela okunelungelo zingathuthukisa khona ngokushesha umkhakha osezingeni eliphezulu. amakhono okuhlasela, okungenzeka kuphazamise isimiso sezinto eziphilayo esibanzi. Abanye baphikisa ngokuthi kungcono ngabavikeli abanekhono ukuhlola nokuqonda lawa maqhinga manje, kunokulinda abadlali abangacophelelanga ukuthi bafinyelele.
Okwengeza ubunzima kule mpikiswano, kube nemibiko yokuthi iMythos isivele ilethe imiphumela emihle kakhulu ezimweni zokuzivikela kuphela. IMozilla, isibonelo, kuthiwa ibone imodeli ibona amakhulu obuthakathaka obungaba khona ku-Firefox ngesikhathi sokuhlolwa kwangaphakathi—izinkinga ezingu-271 kwesinye isibonelo esicashuniwe. Okutholakele okufana nalokhu kusikisela ukuthi izici ezifanayo ezenza i-Mythos ibe namandla okuthuthukisa i-exploit nazo ziyenza ibe umsizi onamandla ngokungavamile wokuqinisa isofthiwe.
I-Anthropic ayizange izibophezele obala kunoma yisiphi isikhathi sokufinyelela okubanzi ku-Mythos, futhi imicimbi efana nokuxhashazwa kwe-M5 cishe izoqinisa izimpikiswano zokugcina noma ngisho nokuqinisa imikhawulo yamanje. Okwamanje, imodeli isalokhu iyithuluzi elikhethekile, elisetshenziswa ezindaweni ezimbalwa zokuphepha ezibaluleke kakhulu lapho uhlobo lwayo lokusetshenziswa okubili luqashelwa ngokusobala.
I-AI ekuphepheni kwe-inthanethi: inkemba esika nhlangothi zombili ekhulayo
Icala lokuxhaphaza laseCalif seliphenduke indawo yokubhekisela engxoxweni ebanzi mayelana Indima ye-AI ekuphepheni kwe-cyberNgakolunye uhlangothi, ikhono lamathuluzi afana ne-Mythos lokuhlaziya ikhodi ngokushesha, ukubona amaphethini amaphutha angabonakali, nokuphakamisa amasu okuhlasela noma okuzivikela kungathuthukisa kakhulu indlela abavikeli abathola futhi balungise ngayo amaphutha ngokushesha. Ngakolunye uhlangothi, amakhono afanayo angafinyeza umjikelezo wokuthuthukiswa kwezenzo eziyinkimbinkimbi, ikakhulukazi uma zisetshenziswa amaqembu anekhono eliphezulu.
Ochwepheshe bezokuphepha bebelokhu bekhathazekile ngekusasa lapho izinhlelo ze-AI zingase zithathe izingxenye ezinkulu zomsebenzi wokuhlasela, kusukela ekuhloleni kuya ekusebenziseni intuthuko kanye nokunyakaza okuseceleni. Isiqephu se-M5 asikameli ukuzenzekela okugcwele, kodwa sinikeza umbono wokuthi kubukeka kanjani lapho I-AI iba yingxenye ebalulekile, enezinga eliphezulu kochungechunge lokuhlasela lwangempela, ngisho nasezandleni zabacwaningi abanomthwalo wemfanelo.
Abaningi emkhakheni manje balindele ukuthi ukuxhashazwa okusizwa yi-AI kuzoba yinto evamile kakhulu ekugcineni kohlu lwezinsongo. Abahlaseli abaxhaswe kahle, kungaba amaqembu obugebengu noma abadlali abasekelwa nguhulumeni, cishe bazohlola noma bakhe izinto zabo ezifanayo namamodeli afana ne-Mythos, bawasebenzise ukuhlunga amakhodi amakhulu futhi bathole amaphuzu abuthakathaka angeke akwaziwa ngabantu ukuwathola bodwa.
Ngesikhathi esifanayo, abavikeli babheka lesi sigameko njengento yokuvusa ilukuluku. Uma ukuhlasela kungasheshiswa kuze kube yileli zinga, khona-ke ukulungisa amapayipi, imikhuba yokubuyekeza ikhodi, kanye nezinqubo zokuphatha ubuthakathaka kungadingeka kucatshangelwe kabusha ukuze kuhambisane nesivinini. Ezinye izinhlangano sezivele zizama amathuluzi okuhlola asizwa yi-AI, zinethemba lokusebenzisa amasu afanayo ukuvimbela ukuhlaselwa ngokuthola nokulungisa ubuthakathaka kuqala.
Ngakho-ke indaba yeCalif-Apple ifika phakathi kwendawo eshintshayo, lapho i-AI ingaseyona inkinga yesikhathi esizayo engaqondakali kodwa ingumdlali osebenzayo enkundleni yezokuphepha. Indlela izinkampani ezikhetha ngayo ukusebenzisa, ukukhawulela, noma ukulawula amamodeli anjalo eminyakeni ezayo cishe ngeke ibumbe nje kuphela izehlakalo ngazinye, kodwa nomzila ophelele wengozi yedijithali.
Uma kuhlanganiswa, ukuxhashazwa ngokumelene ne-Apple's M5 Memory Integrity Enforcement, indima ebalulekile edlalwe yi Ukubuka KukaClaude Mythos, kanye nendlela elawulwayo kodwa ephikisanayo i-Anthropic ephatha ngayo ukufinyelela kumodeli konke kukhomba ohlangothini olufanayo: i-AI ethuthukisiwe ihlanganiswa ngokuqinile nendlela yokuphepha kwe-cyber yesimanje. Leli cala elithile libonisa ukuthi lapho amamodeli olimi asezingeni eliphezulu ebambisana nabacwaningi abanolwazi, bangadalula futhi basebenzise amaphutha amakhulu ezingeni lehadiwe ngezikhathi ezimfushane kakhulu—iqiniso elivelayo okuzodingeka abathengisi nabavikeli bacabange ngalo njengoba beklama isizukulwane esilandelayo sokuvikela.
