Kwenzekeni
I-Trend Micro ithi injini yayo ye-TrendAI ye-agentic exploit-remediation, eqanjwe ngekhodi i-AESIR, ithole u-97% ku- ye-CyberGym futhi yakleliswa kuqala ebhodini labaphambili kusukela ngo-Agasti 2026. Le nkampani ithi lolu hlelo lukhiqize cishe ubufakazi bezinga lebhayithi elingu-1,460 kuwo wonke amaphrojekthi omthombo ovulekile angu-188 namandla aziwayo angu-1,507.
Okuthunyelwe kwebhulogi yezokuphepha ye-Trend Micro yangomhla zingama-26 ku-Agasti, 2026 ithi injini yayo yokulungisa i-TrendAI, ebizwa ngekhodi i-AESIR, ithole amaphuzu angu-97% ku-CyberGym. Le nkampani ichaza i-CyberGym njenge-University of California, Berkeley equkethe ubungozi obuqinisekisiwe obungu-1,507 obuvela kumaphrojekthi amakhulu wemithombo evulekile engu-188. Imisebenzi idinga ukukhiqiza ifayela elithile noma ukulandelana kwebhayithi ephahlaza uhlelo olusengozini kodwa hhayi okuhambisana nalo olupeshisiwe, ngaphandle kokufinyelela kupeshi noma kanambambili elungisiwe ngesikhathi sokuxazulula.
Ngokomthombo, umphumela we-AESIR umele ukuphahlazeka okuhlukene okuqinisekisiwe okukodwa okungaba ngu-1,460. I-Trend Micro ithi uhlelo lwahlolwa ngaphezu kwemizuliswano yentuthuko engu-60, lwasetshenziswa cishe amahora angu-235 e-CPU futhi lwenzeka cishe u-$6,000 ezinkokhelweni ze-AI zohlelo lokuhlela-interface. Inkampani yethula umphumela njengozimele ngokugcwele futhi ithi akekho umuntu owayesethubeni lokuthumela ibhentshimakhi. Lezi zibalo ziyizicelo eziku-akhawunti yenkampani; umthombo awuhlinzeki ngocwaningomabhuku oluzimele noma ifayela lokungena eliphelele.
Uhlelo luchazwa njengephayiphi elineziqu. Iqala ngokuzama ukuhlanganisa kwakudala, ukuguqulwa kwembewu kanye nezindlela zokwakha ezinqumayo, bese ikhuphukela kuma-ejenti aqondiswa yi-AI lapho lezo zindlela zehluleka. I-Trend Micro ithi cishe amaphesenti angu-30 obufakazi bayo ayisebenzisanga izingcingo zemodeli yolimi, kuyilapho ihlanganisa ukucabanga okunwetshiwe kwe-AI mayelana nekota yemisebenzi. Le nkampani ithi ukugijima okungeyona i-AI okuphumelelayo kungathatha imizuzwana engama-60, uma kuqhathaniswa nemizuzu ecishe ibe ngu-18 kanye nezindleko ze-API ezingama-$6 ngomzamo womenzeli we-AI.
I-Trend Micro ibeka ukusebenza okusele ku-ontology yokuba sengozini yangaphakathi equkethe izinkumbulo zesiqephu ezingaphezu kuka-12,500 kanye nezimbewu ezingaphezu kuka-15,500 ezixhaphaza imbewu kumaphrojekthi angaphezu kuka-180. I-ontology ichazwa njengesendlalelo solwazi esisebenzayo esixhumanisa amakilasi okuba sengozini, amafomethi kanambambili, izinhlobo zokuphahlazeka namasu okuguqula izakhi. I-AESIR kubikwa ukuthi isebenzisa amamodeli ayisikhombisa avela ku-Anthropic, Google, OpenAI kanye ne-DeepSeek ezindimeni ezihlukene, ezihlanganisa ukucabanga ngekhodi yomthombo, ukwakhiwa kwefomethi kanambambili kanye nokuhlaziywa kwephrothokholi. Umthombo awuvezi imininingwane eyanele yokusebenzisa ukuze uthole ukuthi ingxenye ngayinye ibe nengxenye engakanani kumphumela wokugcina.
Imininingwane yomthombo: edge.prnewswire.com ↗
Kungani kubalulekile
Umphumela, uma ukhiqizwa kabusha ngokuzimela, ungaphakamisa ukuthi ukuklama kwesistimu—okuhlanganisa ulwazi oluqhubekayo lokuba sengozini, amamodeli amaningi, ukudideka okunqumayo kanye nokubuyekeza okuphikisayo—kungabaluleka ngaphezu kokusebenzisa imodeli eyodwa ehamba phambili. Iphinde ikhombise kokubili amandla kanye nemikhawulo yocwaningo olusizwa yi-AI lokuba sengozini.
Okushiwo okumaphakathi kumayelana nezakhiwo, hhayi nje izinga lamamodeli. I-Trend Micro ithi amasistimu ahamba phambili ku-CyberGym asebenzisa amamodeli amaningi afanayo emngceleni, kuyilapho amaphuzu azo ehluka ngenxa yendlela, inkumbulo, ukusetshenziswa kwamathuluzi nokuqinisekiswa. Ukuqhathanisa kwayo kuklelisa i-AESIR ku-97%, i-Sangfor AI ku-93.2%, iWhitzard yaseFudan University ku-91.2%, i-MDASH ye-Microsoft ku-91% kanye ne-Wiz Atlas ku-90.9%. Ibala i-GPT-5.6 Sol ku-84.5% kanye ne-Claude Mythos 5 ku-83.8%. Lokho kuma kwethulwa yi-Trend Micro njengesimo sebhodi yabaphambili futhi kufanele kuthathwe njengokubikwe ngomthombo kuze kuqinisekiswe i-opharetha yebhentshimakhi.
Emaqenjini okuvikela, isifundo esisebenzayo ukugcina izingcingo zemodeli ezibizayo zamacala ashibhile, amathuluzi abikezelwayo angeke akwazi ukuwaxazulula. Ama-Fuzzer namajeneretha aqondene nefomethi ethile angakwazi ukuphatha amakilasi athile okuba sengozini ngokushesha, kuyilapho amamodeli olimi angasiza ngokuhlaziya ikhodi yomthombo, ukucabanga kokugeleza kokulawula kanye nokwakhiwa kokuxhashazwa kwezinyathelo eziningi. Isendlalelo somzila singanciphisa ukuncika kumhlinzeki oyedwa. Kodwa umnotho ofunwayo uqondene ngqo nohlelo lwe-Trend Micro, umthwalo womsebenzi kanye nesisekelo solwazi oluyimfihlo; akumele zenziwe zibe yisilinganiso sezindleko zomhlaba wonke.
Umthombo uphinde ugcizelele ukubuyekezwa kwamamodeli ahlukene. Kumklamo ochaziwe we-AESIR, imodeli eyodwa iphakamisa inkolelo-mbono yokuxhaphaza, enye imodeli evela kumhlinzeki ohlukile izama ukuyiphikisa bese kuthi owesithathu abe ngabahluleli. Izindima zizungeza imizuliswano. Inzuzo ehlosiwe iwukunciphisa amaphuzu angamanga, njengeziphazamisi ezisobala ezingafinyeleleki, ezivinjwe ukuqinisekiswa okokufaka noma ngaphandle kwemodeli yosongo eshiwo. Lena iphethini engase ibe usizo yamathuluzi okuvikela e-AI, nakuba umthombo unganikezi izilinganiso zamanga, izilinganiso zokuba sengcupheni ezigejiwe noma ukuqhathanisa okulawulwayo ngokumelene nokubuyekezwa kwemodeli eyodwa.
Ibhentshimakhi ibalulekile ngoba ihlola i-artifact ekhonkolo kunempendulo yengxoxo: okokufaka kweleveli ye-byte eziphatha ngendlela ehlukile kusofthiwe esengozini nepheshiwe. Lowo umsebenzi onamandla kunokuveza nje incazelo ezwakalayo yokuba sengozini. Noma kunjalo, ukuphahlazeka okuhlukile kubhentshimakhi elawulwayo ayilingani nokuxhashazwa okuthembekile komhlaba wangempela. Akuqinisekisi ukuthi umhlaseli angakwazi ukufinyelela ikhodi esengozini ekude, adlule ukuzivikela, athole ukufinyelela okuwusizo noma asebenze esikalini. Ngakho-ke umphumela ukhuluma nekhono lokwakha lokuxhaphaza ngaphansi kwezimo zokulinganiswa, hhayi kubo bonke ubungozi be-inthanethi noma ukulungela ukuhlasela okuzimele.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ongakubuka ngokulandelayo
Imibuzo ebalulekile ukuthi i-CyberGym noma abacwaningi abazimele bayawaqinisekisa yini amaphuzu, noma ukuhlola kungenziwa kabusha, nokuthi ungakanani umphumela uncike kukhorasi eyimfihlo ye-Trend Micro yokuba sengozini. Ibhentshimakhi futhi ayibonisi ukuthi isistimu ithola ukuhlaselwa okusebenzayo, isebenza kuyo yonke isofthiwe engaziwa noma ingasebenza ngokuphephile ngaphandle kokubuyekezwa komuntu.
Okokuqala, ukuqinisekiswa okuzimele kubalulekile. Umthombo uthi i-CyberGym inebhodi yabaphambili esemthethweni kanye nesistimu yokuhambisa ezimele, kodwa ayibandakanyi irekhodi lokuma eliqondile, ikhodi ekwazi ukukhiqiza kabusha, isethi yobufakazi ethunyelwe noma iphrothokholi enemininingwane yokuhlola amaphuzu enkampani. Ukuqinisekisa kufanele kusungule inguqulo eqondile yesilinganiso, imithetho yokufakwa komsebenzi, ihadiwe nemikhawulo yesikhathi, izinguqulo zamamodeli, ukufinyelela kwamathuluzi kanye nokuthi wonke amasistimu ahlolwe ngaphansi kwezimo ezifanayo.
Okwesibili, abacwaningi kufanele bahlole inzuzo yedatha yangasese. I-Trend Micro ithi i-ontology yayo ihlanganisa iminyaka engaphezu kwamashumi amabili yocwaningo lokuba sengozini oluhlobene ne-Trend Micro Research kanye ne-Zero Day Initiative, okuhlanganisa izinkulungwane zembewu yokuxhashazwa kanye nemiphumela yokuzibandakanya. Lolo lwazi oluqoqiwe lokusebenza lungase lubaluleke, kodwa futhi kwenza umphumela ube nzima ukuqhathanisa nesistimu esebenzisa ulwazi lomphakathi kuphela. Okubalulekile okungaziwa ukuthi indlela ye-AESIR idlulisela kumaphrojekthi amasha, amakilasi okuba sengcupheni namafomethi esofthiwe awekho kukhorasi yawo yomlando.
Okwesithathu, izimangalo zokuthunyelwa zidinga ubufakazi obuhlukene. Ibhulogi ixoxa ngengxenye yokuzingela usongo engahlola ukuthunjwa kwebhodwe lezinyosi futhi ihlonze ukuxhashazwa okungase kube khona kwengqalasizinda ye-AI, kodwa lowo msebenzi awufani nomphumela we-CyberGym. Ibhentshimakhi ayilinganisi ukuthi i-AESIR ithola imikhankaso esebenzayo, ihlukanisa umsebenzi onobungozi ekuhloleni okungenabungozi noma ihlola ukuthi ukulungisa okuthunyelwayo kususe ubungozi bomhlaba wangempela. Leyo mibuzo izodinga idatha yokusebenza, ukudalulwa kwezindlela zokuhlola nokuvikela ngokucophelela.
Okokugcina, amaqembu ezokuphepha kufanele abheke ukuthi isistimu ikusingatha kanjani ukungaqiniseki nokulawula komuntu. I-Trend Micro ithi imisebenzi esele enzima kakhulu ihilela ukulandelana okunembayo kokukhishwa kwamakhodi kwe-FFmpeg, imishini yesifunda ye-Ghostscript kanye nezakhiwo ze-smartcard ASN.1. Iphinde ixwayise ngokuthi amasistimu amamodeli amaningi angahlushwa ukukhathala kwesabelo, ukwehluleka kwabahlinzeki buthule, ukweqiwa kwezindleko kanye nokuhlehla okufihliwe kwemodeli ngayinye. Ngaphambi kokuthi amasistimu anjalo asetshenziswe ekukhiqizeni, amaqembu adinga ukugawulwa kwemithi okuzimele, amamethrikhi engxenye ngayinye, izimvume ezilawulwayo, ukubuyekezwa komuntu ngezinqumo zokuxhashazwa kanye nobufakazi bokuthi ukuhlola okuphikisanayo kunciphisa amaphuzu angamanga ayingozi ngaphandle kokucindezela okutholakele kwangempela.