Kalmomin sirri sun ci gaba da zama babbar hanyar tabbatar da ainihi duk da sanannun raunin tsaro. Masu amfani suna son ƙirƙirar kalmomin sirri bisa ga tsarin da ake iya hasashe, wanda ke sa su zama masu rauni ga hare-haren zato. Ba za a iya ƙididdige tsaron irin waɗannan tsare-tsaren ta hanyar sigogin sirri na al'ada ba amma yana buƙatar ingantaccen tsarin halayyar abokin gaba. Wannan takarda ta magance wani gibi mai mahimmanci: babban karkatar ma'auni da aka gabatar lokacin da masu bincike suka yi amfani da hare-haren ƙamus na kasuwa, waɗanda ba su kama dabarun abokan gaba na ainihi masu sauyi da ƙwarewa ba.
Masu fasa kalmomin sirri na ainihi suna amfani da hare-haren ƙamus masu inganci da yawan aiki tare da ƙa'idodin gyara (misali, ta amfani da kayan aiki kamar Hashcat ko John the Ripper). Ingancin waɗannan hare-haren ya dogara da saitunan da aka daidaita da ƙwarewa—takamaiman nau'ikan jerin kalmomi da ƙa'idodi—waɗanda aka ƙirƙira ta hanyar gogewa. Binciken tsaro da ya dogara da saitunan asali yana ƙara ƙima da yawa ga ƙarfin kalmar sirri, yana gabatar da karkatar ma'auni wanda ke lalata ingancin sakamakon tsaro.
Babbar matsala ita ce rarrabuwar kawuna tsakanin ƙirar kalmomin sirri na ilimi da ayyukan fasa sirri na ainihi. Bincike kamar na Ur et al. (2017) ya nuna cewa ma'aunin ƙarfin kalmar sirri yana da matuƙar mahimmanci ga tsarin abokin gaba da aka yi amfani da shi. Yin amfani da ƙirar mara ƙarfi ko gama gari yana haifar da ƙarin ƙimar tsaro, yana haifar da tunanin aminci na ƙarya.
Hare-haren ƙamus na al'ada suna tsaye. Suna amfani da ƙayyadaddun tsarin ƙa'idodin gyara (misali, magana ta leet, ƙara lambobi a ƙarshe) zuwa ga ƙayyadaddun jerin kalmomi a cikin tsari da aka ƙayyade. Ba su da dacewar ƙwararrun mutane waɗanda za su iya:
Marubutan sun ba da shawarar hanya mai fuska biyu don sarrafa dabarun zato kamar na ƙwararru, don rage dogaro ga saitun hannu da ilimin yanki.
An horar da cibiyar sadarwar jijiyoyi mai zurfi (DNN) don ƙirar rarraba yuwuwar kalmomin sirri. Babban ƙirƙira shine horar da wannan ƙirar ba kawai akan bayanan kalmomin sirri na danye ba, amma akan jerin ƙa'idodin gyara da ƙwararrun masu fasa sirri suka yi amfani da su ga kalmomin tushe. Wannan yana ba da damar DNN ta koyi "ƙwarewar" abokin gaba—madaidaicin canje-canje da tsarinsu mai inganci.
Maimakon ƙayyadaddun tsarin ƙa'idodi, harin yana amfani da dabarun zato masu sauyi. DNN tana jagorantar samar da kalmomin sirri masu yuwuwa ta hanyar aiwatar da canje-canje a jere tare da yuwuwar da aka ƙayyade bisa ga halin yanzu na kalma da mahallin hari. Wannan yana kwaikwayon ikon ƙwararren da ya dace da hanyar harin a ainihin lokaci.
Ana iya ɗaukar tsarin a matsayin mai samar da yuwuwar. Idan aka ba da kalmar tushe $w_0$ daga ƙamus, ƙirar tana samar da kalmar sirri $p$ ta hanyar jerin $T$ na canje-canje (ƙa'idodin gyara $r_t$). An ƙirƙiri yuwuwar kalmar sirri kamar haka: $$P(p) = \sum_{w_0, r_{1:T}} P(w_0) \prod_{t=1}^{T} P(r_t | w_0, r_{1:t-1})$$ inda $P(r_t | w_0, r_{1:t-1})$ shine yuwuwar aiwatar da ƙa'idar $r_t$ idan aka ba da kalmar farko da tarihin ƙa'idodin da suka gabata, kamar yadda DNN ta fitar. Wannan tsari yana ba da damar aiwatar da ƙa'idodin da suka sani da mahallin, waɗanda ba su da layi.
An gudanar da gwaje-gwaje akan bayanan kalmomin sirri na ainihi masu girma da yawa (misali, RockYou, LinkedIn). An kwatanta ƙirar da aka tsara da ƙirar kalmomin sirri na yuwuwar zamani (misali, ƙirar Markov, PCFGs) da hare-haren ƙamus na yau da kullun tare da tsarin ƙa'idodi na shahara (misali, best64.rule, d3ad0ne.rule).
Babban ma'auni shine lambar zato—nawa ne ake buƙatar zato don fasa kashi na musamman na kalmomin sirri. Sakamakon ya nuna cewa harin ƙamus mai sauyi wanda DNN ke ɗaukar nauyi:
Bayanin Chati: Chati mai layi zai nuna jimillar kashi na kalmomin sirri da aka fasa (Y-axis) akan lissafin lambar zato (X-axis). Tsarin hanyar da aka tsara zai tashi da sauri da girma fiye da na PCFG, Markov, da hare-haren ƙamus masu tsayayye, musamman a cikin matsayin zato na farko (misali, zato na farko 10^9).
Takardar ta ƙididdige raguwar karkatar ma'auni. Lokacin tantance ƙarfin manufar kalmar sirri, yin amfani da harin tsaye zai iya yanke shawarar cewa 50% na kalmomin sirri suna tsayayya da zato 10^12. Harin mai sauyi da aka tsara, wanda ke ƙirar abokin gaba mai ƙarfi, zai iya nuna cewa 50% an fasa su ta hanyar zato 10^10—wanda ke nuna ƙarin ƙima sau 100 ta ƙirar tsaye. Wannan yana nuna mahimmancin mahimmanci na ingantaccen tsarin abokin gaba don yanke shawara kan manufofi.
Hali: Ƙungiyar tsaro tana son tantance juriyar kalmomin sirri na masu amfani da su a kan harin da aka yi niyya da ƙwarewa.
Hanyar Al'ada (Mai Karkata): Suna gudanar da Hashcat tare da jerin kalmomin rockyou.txt da tsarin ƙa'idodin best64.rule. Rahoton ya ce: "80% na kalmomin sirri za su tsira daga zato biliyan 1."
Tsarin da aka Tsara (Mai Rage Karkatarwa):
Fahimta ta Asali: Wannan takarda ta kai hari mai ma'ana kan aibi mai yaduwa amma sau da yawa ana watsi da shi a cikin binciken tsaron yanar gizo: karkatar "gibin ƙwarewa". Shekaru da yawa, kimanta ƙarfin kalmar sirri na ilimi an gina shi akan yashi—ta amfani da ƙirar abokan gaba masu sauƙi, masu tsayayye waɗanda ba su da kamanni da ƙwararrun mutane masu dacewa da kayan aiki a cikin daji. Pasquini et al. ba kawai suna ba da algorithm mafi kyau ba; suna tilasta wa fagen fuskantar makafinsa na hanyoyin bincike. Babban nasara shine tsara matsalar ba a matsayin "fasa kalmar sirri mafi kyau" ba amma a matsayin "simulation na abokin gaba mafi kyau," canji mai mahimmanci amma mai mahimmanci a cikin hangen nesa mai kama da motsi daga masu rarrabe masu sauƙi zuwa Nets na Abokin Gaba na Samarwa (GANs) a cikin AI, inda ingancin mai samarwa ya bayyana ta ikonsa na yaudarar mai nuna bambanci.
Kwararar Ma'ana: Hujja tana da layi mai jan hankali. 1) Barazana ta ainihi = hare-hare masu sauyi da aka daidaita da ƙwarewa. 2) Al'adar bincike ta gama gari = hare-hare masu tsayayye, na kasuwa. 3) Saboda haka, akwai babban karkatar ma'auni. 4) Magani: Sarrafa saitunan ƙwararru da dacewa ta amfani da AI. Amfani da DNN don ƙirar jerin ƙa'idodi yana da kyau. Ya gane cewa ilimin ƙwararru ba kawai jakar ƙa'idodi ba ne, amma tsarin yuwuwar—nahawun fasa sirri. Wannan ya yi daidai da nasarar ƙirar jerin kamar Transformers a cikin NLP, yana nuna cewa marubutan suna amfani da darussan daga fagagen AI na kusa yadda ya kamata.
Ƙarfi & Aibobi: Babban ƙarfi shine tasiri mai amfani. Wannan aikin yana da amfani nan take ga masu gwajin shiga da masu duba tsaro. Hanyarsa ta tushen DNN kuma tana da inganci a cikin koyon rikitattun tsarin fiye da tsoffin hanyoyin PCFG. Duk da haka, wani babban aibi yana ƙoƙari a cikin dogaro da bayanan horo. "Ƙwarewar" ƙirar an koya ta daga halayen ƙwararru da aka lura (jerin ƙa'idodi). Idan bayanan horo sun fito ne daga wata al'umma ta musamman na masu fasa sirri (misali, waɗanda ke amfani da Hashcat ta wata hanya), ƙirar na iya gaji karkatansu kuma ta rasa sabbin dabarun. Wani nau'i ne na kwaikwayo, ba hankali na dabarun gaskiya ba. Bugu da ƙari, kamar yadda aka lura a cikin wallafe-wallafen koyo na tarayya (misali, aikin Google AI), tasirin sirri na tattara irin waɗannan bayanan "binciken hari" masu mahimmanci don horo ba su da ƙarami kuma ba a bincika su sosai.
Fahimta Mai Aiki: Ga masu aiki a masana'antu: Daina amfani da tsarin ƙa'idodi na asali don tantance haɗari. Haɗa ƙirar masu sauyi, masu sanin mahalli kamar wannan a cikin hanyoyin gwajin tsaron ku. Ga masu bincike: Wannan takarda ta kafa sabon ma'auni. Dole ne ƙirar kalmomin sirri na gaba a tabbatar da su a kan abokan gaba masu dacewa, ba masu tsayayye ba. Gaba gaba shine rufe madauki—ƙirƙirar masu tsaron AI waɗanda za su iya tsara kalmomin sirri ko manufofin da suka ƙarfi a kan waɗannan hare-haren masu sauyi na AI, suna matsawa zuwa ga tsarin haɗin gwiwar abokan gaba mai kama da GANs, inda ƙirar maharin da mai tsaro suka inganta tare. Zamanin tantance kalmomin sirri a cikin sarari mai tsayayye ya ƙare, ko ya kamata ya ƙare.