Ukusetshenziswa kwezikhali zamathuluzi e-AI akhiqizayo afana ne-ChatGPT leyo wonke umuntu kade elindile kancane, kancane kancane eqala ukuthatha ifomu. Emiphakathini eku-inthanethi, amakati anelukuluku abambisana ezindleleni ezintsha zokwephula imithetho yokuziphatha ye-ChatGPT, evame ukwaziwa ngokuthi “i-jailbreaking,” futhi izigebengu ze-inthanethi zakha inethiwekhi yamathuluzi amasha ukuze kuthuthukiswe noma adale amamodeli ezilimi amakhulu (LLM) ukuze afeze izinjongo ezinonya.
Njengoba nje ingaphezulu komhlaba, i-ChatGPT ibonakala igqugquzele isiphithiphithi ezinkundleni ezingaphansi komhlaba. Kusukela ngoDisemba, abaduni bebelokhu befuna okusha nokuqanjiwe icela ukukhohlisa i-ChatGPT, Futhi ama-LLM omthombo ovulekile angaphinda awenze ngenjongo yezinjongo ezinonya.
Umphumela, ngokusho kwebhulogi entsha evela ku-SlashNext, usengumphakathi osafufusa kodwa ochumayo wokugebenga we-LLM, onezinto eziningi ezihlakaniphile kodwa ama-malware ambalwa anikwe amandla i-AI afanele ukucabanga okwesibili.
Lokho Abaduni Abakwenzayo Nge-AI LLMs
Ubunjiniyela obusheshayo kuhlanganisa ukubuza ama-chatbot ngobuhlakani njengemibuzo ye-ChatGPT ehloselwe ukuwakhohlisa, ukuwenza ephule imithetho yawo ehleliwe ngokumelene, sisho, ukudala uhlelo olungayilungele ikhompuyutha, ngaphandle kokuthi amamodeli akwazi. Lokhu kuwukusebenzisa amandla anonya, kuchaza uPatrick Harr, oyi-CEO ye-SlashNext: “Abaduni bazama nje ukubheka ezindaweni zonogada. Iyini imiphetho? Ngimane ngishintsha njalo imiyalelo, ngiyibuze ngezindlela ezahlukene ngenze engikufunayo.”
Ngoba kuwumsebenzi okhathazayo, futhi ngenxa yokuthi wonke umuntu uhlasela impokophelo efanayo, kungokwemvelo ukuthi imiphakathi ye-inthanethi enosayizi onempilo yakheke kulo mkhuba ukuze wabelane ngamathiphu namasu. Amalungu ale miphakathi ehlakazekile ejele aklwebhana emhlane, asizana ukwenza i-ChatGPT iqhekeke futhi yenze izinto onjiniyela ababehlose ukuyivimba ukuthi izenze.
Onjiniyela abasheshayo bangakwazi ukuzuza okuningi ngokudlalwa kwamagama okuhle, noma kunjalo, uma i-chatbot okukhulunywa ngayo yakhiwe ngendlela eqinile njengoba injalo i-ChatGPT. Ngakho-ke, okukhathaza kakhulu ukuthi abathuthukisi bohlelo olungayilungele ikhompuyutha sebeqala ukuhlela ama-LLM ngokweziphetho zabo ezimbi.
Usongo Oluzayo Lwe-WormGPT nama-LLM anonya
Kwabizwa umnikelo I-WormGPT uvele ngoJulayi ukuze aqale lolu hlelo olubi lwe-LLM. Kungenye indlela yesigqoko esimnyama kumamodeli e-GPT aklanyelwe ngokukhethekile imisebenzi enonya njenge-BEC, uhlelo olungayilungele ikhompuyutha, nokuhlasela kobugebengu bokweba imininingwane ebucayi, imakethwa ezinkundleni eziyimfihlo “njenge-ChatGPT kodwa [ngaphandle] kwemingcele yokuziphatha noma imikhawulo.” Umdali we-WormGPT uthe uyakhele phezu kwemodeli yolimi ngokwezifiso, eqeqeshwe emithonjeni ehlukahlukene yedatha, egcizelela idatha ehlobene nokuhlasela ku-inthanethi.
“Kusho ukuthini kubaduni,” kuchaza u-Harr, “manje sengingakwazi ukuthatha, ngithi, ukuyekethisa kwe-imeyili yebhizinisi (BEC), noma ukuhlasela kobugebengu bokweba imininingwane ebucayi, noma ukuhlasela kwe-malware, futhi ngikwenze lokhu ngezikali ngezindleko ezincane kakhulu. Futhi ngingaqondiswa kakhulu kunakuqala.”
Kusukela nge-WormGPT, inani lemikhiqizo efanayo liye lahlanganiswa emiphakathini ye-inthanethi enomthunzi, kuhlanganise FraudGPT, ekhangiswa “njenge-bot engenamingcele, imithetho, [kanye] nemingcele” ngumdlali osongelayo othi ungumdayisi oqinisekisiwe ezimakethe ezihlukahlukene ezingaphansi komhlaba ezimnyama zeWebhu, okuhlanganisa i-Empire, WHM, Torrez, World, AlphaBay, kanye ne-Versus. Futhi u-August waletha ukubonakala kwethempeli I-DarkBART kanye ne-DarkBERT yama-cybercriminal chatbots, esekelwe ku-Google Bard, abacwaningi ngaleso sikhathi ababethi imele i-AI ephikisanayo, okuhlanganisa ukuhlanganiswa kwe-Google Lens yezithombe kanye nokufinyelela okusheshayo kuso sonke isisekelo solwazi lwe-cyber-underground.
Ngokusho kwe-SlashNext, lezi ziyanda manje, iningi lazo lakhela phezu kwamamodeli omthombo ovulekile njenge-OpenAI's OpenGPT. Inqwaba yabaduni abanamakhono aphansi bavele bayenze ngendlela oyifisayo, bayifihle ngesonga, bese beshaya igama elisabekayo “___GPT” kuyo (isb. “BadGPT,” “DarkGPT”). Ngisho nale minikelo ye-ersatz inendawo yayo emphakathini, nakuba, inikeza imikhawulo embalwa kanye nokungaziwa ngokuphelele kwabasebenzisi.
Ukuvikela Ku-Next-Gen AI Cyberweapons
I-WormGPT, noma inzalo yayo, noma onjiniyela abasheshayo, abethuli ingozi enkulu kangaka emabhizinisini okwamanje, ngokusho kwe-SlashNext. Noma kunjalo, ukwanda kwezimakethe ezigqekeza ejele kusho ukuthi amathuluzi amaningi ayatholwa yizigebengu zamakhompiyutha, okubuye kuvezwe ushintsho olubanzi kwezobunjiniyela bezenhlalo, kanye nendlela esivikela ngayo kubo.
UHarr uyeluleka: “Unganciki ekuqeqesheni, ngoba lokhu kuhlasela kuqondile, kuqondile, futhi kuqondiswe kakhulu, kakhulu kunalokho okwakunjalo ngaphambili.”
Kunalokho, ubhalisela umbono okuvunyelwene ngawo ngokuvamile wokuthi izinsongo ze-AI zidinga ukuvikelwa kwe-AI. “Uma ungenawo amathuluzi e-AI okuthola nokubikezela nokuvimbela lezi zinsongo, uzobe ungaphandle ubheka ngaphakathi,” usho kanje.
- I-SEO Powered Content & PR Distribution. Khuliswa Namuhla.
- I-PlatoData.Network Vertical Generative Ai. Zinike Amandla. Finyelela Lapha.
- I-PlatoAiStream. I-Web3 Intelligence. Ulwazi Lukhulisiwe. Finyelela Lapha.
- I-PlatoESG. Izimoto / EVs, Ikhabhoni, I-CleanTech, Amandla, Environment, Ilanga, Ukuphathwa Kwemfucuza. Finyelela Lapha.
- I-PlatoHealth. I-Biotech kanye ne-Clinical Trials Intelligence. Finyelela Lapha.
- I-ChartPrime. Phakamisa umdlalo wakho wokuhweba ngeChartPrime. Finyelela Lapha.
- I-BlockOffsets. Ukwenza kube yisimanje Ubunikazi be-Environmental Offset. Finyelela Lapha.
- Source: https://www.darkreading.com/application-security/chatgpt-jailbreaking-forums-dark-web-communities
