Out of Scope Ba Fita Daga Haɗari Ba: Sabuntawar Model Risk ta 2026 Ta Bar Samfuran da Bankuna Ke Aiwatarwa

An sake rubuta koyarwar model risk ta shekara goma sha biyar a watan Afrilu, kuma sakin layi da ya fi nauyi shi ne wanda ya faɗi abin da guidance ba ta rufe ba. A ranar 17 ga Afrilu 2026 Federal Reserve, OCC da FDIC sun fitar da guidance da aka sabunta kan model risk management, wadda ta maye gurbin tsarin 2011 da masana'antar ta shirya kanta a kansa tsawon shekaru goma sha biyar. Sabuntawar tana da tsanaki, tana kan haɗari, kuma galibi abin maraba ce. Ta kuma faɗa a fili cewa samfuran generative AI da agentic AI ba sa cikin scope ɗinta — tare da request for information da za ta biyo baya. Saboda haka bankuna suna cikin wani lokaci na tsaka mai wuya: suna aiwatar da samfuran da tsarin ya cire, ƙarƙashin tsarin da ba a rubuta ba tukuna.

Taƙaitawa ga Manyan Jami'ai

An yi ritayar yarjejeniyar 2011. SR 26-2 da OCC Bulletin 2026-13, dukkansu masu kwanan wata 17 ga Afrilu 2026, sun maye gurbin guidance ta 2011 da abokiyar ta ta 2021. Gine-ginen ginshiƙai uku — ci gaba mai kyau, effective challenge, ongoing monitoring — sun ci gaba.

Proportionality yanzu a bayyane take. Hukumomin sun ce ana sa ran guidance ɗin za ta fi dacewa da ƙungiyoyin banki masu sama da dala biliyan 30 na jimlar dukiya, yayin da take ci gaba da dacewa da ƙananan cibiyoyi masu model risk mai muhimmanci. Wannan canjin nauyi ne mai ma'ana, ba keɓancewa ba.

An jinkirta tambayar AI, ba a amsa ta ba. An cire generative da agentic AI daga wannan guidance, tare da request for information da za ta biyo baya wadda ta shafi model risk management gaba ɗaya da kuma amfanin bankuna da AI musamman.

Jinkiri ba izini ba ne. Karatun daidai shi ne cewa hukumomin ba su yanke shawara ba tukuna kan yadda za su kula da waɗannan samfuran — abin da ya sha bamban sosai da yanke shawarar cewa ba sa buƙatar shugabanci.

Abin da Hukumomin Suka Canza da Gaske

Cire sharhi a gefe, sabuntawar Afrilu tana yin abubuwa uku.

Tana yin ritayar yarjejeniyar 2011. Guidance da aka sabunta ta maye gurbin guidance kan model risk da aka fitar a 2011 da abokiyar ta ta 2021, tana rufe babi na shekaru goma sha biyar wanda a lokacinsa SR 11-7 ta zama ma'auni na duniya a aikace ga shugabancin samfuri fiye da cibiyoyin da ta ɗaure.

Tana sanya proportionality a bayyane. Ana sa ran guidance ɗin za ta fi dacewa da ƙungiyoyin banki masu sama da dala biliyan 30 na jimlar dukiya, kuma tana ci gaba da dacewa da ƙananan ƙungiyoyi masu babban fallasa ga model risk saboda yawa da rikitarwar samfuransu ko ayyukansu a wajen bankin al'umma na gargajiya. Ƙwarewa ya kamata ta bi girma, rikitarwa da bayanin haɗari.

Tana riƙe gine-ginen. Conceptual soundness a ci gaba, effective challenge ta hanyar bita mai zaman kanta, da ongoing monitoring bisa ƙofofin da aka rubuta duk sun tsira. Cibiyoyin da suka gina da kyau bisa tsarin 2011 ba sa farawa daga sifili.

Tebur 1: Abin da ya canza, da abin da ya ci gaba

Sinadari	Tsarin 2011	Guidance da aka sabunta 2026
Matsayi	SR 11-7 da abokiyar ta ta 2021	An maye gurbinsu, 17 ga Afrilu 2026
Gine-ginen tushe	Ci gaba, tantancewa, shugabanci	Ya ci gaba kusan gaba ɗaya
Proportionality	A ɓoye, ana amfani da shi ta hanyar kulawa	A bayyane: an daidaita shi da girma, rikitarwa da bayanin haɗari
Dacewar da aka bayyana	Mai faɗi	Ya fi dacewa sama da dala biliyan 30 na jimlar dukiya; har yanzu yana dacewa a ƙasa inda model risk yake da muhimmanci
Generative da agentic AI	Sun rigaye tambayar	A fili out of scope, tare da RFI da za ta biyo baya

Jimlar da Take da Muhimmanci

Yawancin sabuntawar ci gaba ne. Sakin layi guda ba haka yake ba:

Samfuran generative AI da agentic AI sabbi ne kuma suna canzawa cikin sauri. Saboda haka, ba sa cikin scope na wannan guidance.

Idan aka karanta tare da alƙawarin fitar da request for information da ta shafi model risk management gaba ɗaya, da amfanin bankuna da AI ciki har da generative da agentic AI musamman, ma'anar ta bayyana sosai. Hukumomin ba sa cewa waɗannan tsarukan suna da ƙarancin haɗari. Suna cewa ba su warware yadda za su kula da su ba, kuma za su fi son yin shawarwari maimakon kafa tsarin da fasaha mai gudu ke wuce shi cikin zagaye guda.

Wannan sana'ar tsara ƙa'ida ce mai kariya. Amma kuma matsayi ne mai wuya ga banki da ya riga ya sanya generative AI a aiki.

Domin tambayar aiki da kwamitin haɗari yake yi ba "shin wannan yana cikin scope na guidance ta Afrilu?" ba ce. Ita ce "idan wannan ya lalace, me za a tambaye mu?" Kuma amsar tambaya ta biyu ba ta canza ba.

Fita Daga Scope Ba Fita Daga Haɗari Ba

Kuskuren karatu mafi tsada a 2026 shi ne cewa cirewa daga guidance kan model risk yana nufin cirewa daga kulawa. Ba haka ba ne. Samfurin da ya faɗi wajen wannan tsari na musamman har yanzu yana zaune daidai cikin wasu da dama.

Tebur 2: Inda tsarin GenAI da ke out of scope har yanzu yake faɗawa

Samfurin yana yin wannan	Tsarin da har yanzu yake amsawa gare shi	Tambayar da za a yi maka
Yana gudana kan samfurin tushe na mai kaya	Third-party risk da haɗarin ba da aiki waje	Za ka iya tabbatar da bincike mai zurfi, shirin fita da nazarin taruwa kan mai bayarwa?
Yana taɓa shawarar bashi ko farashi	Kariyar mabukaci da fair lending	Za ka iya nuna cewa sakamakon ba wariya ba ne, kuma ka sake gina dalilin da ya sa aka kai ga shi?
Yana zaune a cikin tsarin da ke fuskantar abokin ciniki ko na aiki	Haɗarin aiki da juriya	Me zai faru idan ya lalace, kuma wa zai lura?
Yana samar da sakamakon da ma'aikata suka dogara da shi	Shugabanci da alhakin manyan jami'ai	Wa ya mallaki wannan tsarin, kuma me aka gaya masa kan iyakokinsa?
An aiwatar da shi a EU ko Birtaniya	EU AI Act; PRA SS1/23	Amsar scope a nan ta bambanta — duba ƙasa

Babu wani daga cikin waɗannan da ke sabon wajibi ba. Wannan shi ne ainihin abin: guidance ta Afrilu ta rage abin da ita take rufewa, kuma ba ta cire komai daga kowane wani ba.

Akwai kuma rashin daidaito a supervisory expectations da ya cancanci a faɗe shi a fili. Idan tsari ya yi shiru, masu binciken ba sa yin shiru. Suna komawa ga tsammanin aminci da ƙarfi na gaba ɗaya, waɗanda suka fi faɗi kuma ba a iya hasashen su kamar ma'aunin model risk da aka rubuta. Fita daga scope na wata ƙa'ida ta musamman sau da yawa yana nufin a auna ka bisa wadda ta fi rashin bayyana.

Rarrabuwar Bangarorin Tekun Atlantika da Bankin Ƙungiya Zai Gudanar

Ga cibiyar da ke aiki a bangarorin biyu na Tekun Atlantika, shawarar Afrilu ta haifar da rarrabuwa mai rai.

Sanarwar kulawa ta Birtaniya kan ka'idojin model risk management ga bankuna, SS1/23, ta fara aiki a 17 ga Mayu 2024. An gina ta kan ka'idoji biyar — ganowa da model inventory, shugabanci, ci gaba da aiwatarwa da amfani, tantancewa mai zaman kanta, da matakan rage haɗari ga samfuran da ke da gazawa — kuma tana tsaka-tsaki kan fasaha da gangan. Ba a keɓe hankali na wucin gadi da machine learning ba; an tsammace su.

Don haka samfuri guda na generative, wanda aka aiwatar a ƙungiya ɗaya, a layin kasuwanci ɗaya, yana cikin iyakar model risk na Birtaniya kuma yana wajen wanda Amurka ta sabunta. Ƙara wajibin EU AI Act ga aiwatarwa a Turai kuma amsoshi uku ne, ba biyu ba.

Cibiyoyi suna da amsoshi biyu kaɗai masu ma'ana.

Gudanar da tsari guda bisa ma'auni mafi tsanani. Yi shugabancin generative AI kamar yana cikin scope a ko'ina, domin yana cikin scope a wani wuri kuma zai kasance cikin scope da faɗi bayan RFI ta warware. Wannan yana kashe kuɗi da yawa yanzu kuma yana da kariya a kowace hurumi.

Gudanar da iyakoki daban-daban da gaske. Yana yiwuwa a doka, amma yana da wahala a aiki. Yana nufin lissafi biyu, ma'aunin tantancewa biyu da tarin shaida biyu ga samfuri guda — da fallasa ga suna wajen bayyanawa, bayan wani lamari, cewa nau'in kula da zai kama shi shi ne wanda aka yi amfani da shi a ɗayan hurumin.

Na farko shi ne amsar da ta dace ga kusan kowace cibiya mai girman da ya sa wannan ya zama tambaya. Na biyu shi ne abin da ke faruwa ta atomatik idan babu wanda ya yanke shawara.

Abin da Za a Yi da Lokacin Tsaka Mai Wuya

RFI tana zuwa. Taga tsakanin yanzu da tsarin da aka warware shi ne ɓangaren da dole a sarrafa shi da gangan, domin shi ne lokacin da aiwatarwa ke ci gaba amma takardu ba sa ci gaba.

Alƙawura uku sun sa wannan lokacin ya zama abin jurewa.

Yi lissafi yanzu, rarrabawa daga baya. Kayan aikin da ya fi daraja shi ne cikakken rajistar inda tsarukan generative da agentic suke gudana da gaske — ciki har da waɗanda aka saya a matsayin fasali cikin kayan SaaS, inda yawancinsu ke ɓoyewa. Ko kowanne a hukumance "samfuri" ne ƙarƙashin wani ma'ana na gaba tambaya ce da za ka iya amsa da zarar ka san suna nan. Ba za ka iya rarraba abin da ba ka samu ba.

Rubuta hanyar tantancewa da za ka kare. Backtesting na gargajiya ba ya canjawa da kyau zuwa tsarin da ba shi da tabbas, wanda shi ne ainihin dalilin da ya sa hukumomin suka jinkirta. Wannan ba dalili ba ne na rashin samun kowace hanya. Gwajin samarwa bisa lamurran da aka ajiye a gefe, bin diddigin sauyawar da mutum ya yi, nazarin sakamako a layi ɗaya bisa tsarin da ake amfani da shi, da ƙofofin karɓa da aka rubuta duk suna nan yau kuma duk mai kulawa zai iya karanta su.

Amsa RFI. Cibiyar da ta gina hanyar aiki a wannan lokacin tana da wani abu na zahiri da za ta faɗa, kuma shawarwarin shi ne inda ake tsara ma'aunin ƙarshe. Kamfanonin da suka yi shiru suna gadon duk abin da masu magana suka yi ciniki a kai.

Littafin Aiki

Sake auna kanka bisa guidance da aka sabunta yanzu. Tabbatar da waɗanne daga cikin samfuranka na yanzu suka koma ƙarƙashin harshen proportionality a bayyane, a bangarori biyu. Wasu za su buƙaci ƙasa; wasu za su buƙaci ƙari.
Kada ka cire komai daga model inventory saboda cirewar AI. Cirewa shawara ce da za ta yi mummunan karatu a nan gaba, kuma sake ganowa yana kashe fiye da riƙewa.
Yi wa kowane tsarin generative da agentic alama a lissafin a matsayin mai jiran rarrabawa. Sanya lokacin tsaka mai wuya ya bayyana a rajista maimakon ya ɓace daga cikinta.
Haɗa kowanne da tsarukan da yake amsawa gare su — third-party, aiki, kariyar mabukaci, fair lending — kuma tabbatar da kowanne yana da mai shi da aka ambaci sunansa ƙarƙashin waɗannan.
Karɓi ma'auni mafi tsanani na bangarorin biyu na Tekun Atlantika a matsayin manufar ƙungiya. Lissafi guda, ma'aunin tantancewa guda, tarin shaida guda.
Shirya amsar RFI. Ita ce hanya mafi arha ta yin tasiri kan ma'aunin da za a bincike ka bisa shi.

Hukumomin sun yi wani abu mai ma'ana a Afrilu: sun ƙi rubuta ƙa'ida ga fasahar da ba su fahimta sosai don su iyakance ta ba. Aikin da ya dace da haka ga banki shi ne ya ƙi ɗaukar hakan a matsayin izini.

Tambayoyin da Ake Yawan Yi

Shin guidance ta Afrilu 2026 tana nufin babu ƙa'ida kan generative AI a bankin Amurka?
A'a. Tana nufin samfuran generative da agentic AI suna wajen scope na wannan guidance ta musamman. Waɗannan tsarukan sun ci gaba da kasancewa ƙarƙashin third-party risk management, haɗarin aiki, kariyar mabukaci, fair lending da tsammanin aminci da ƙarfi na gaba ɗaya. Hukumomin sun kuma ɗaura alƙawarin request for information da ta shafi amfanin bankuna da AI, ciki har da generative da agentic AI.

Me daidai aka maye gurbinsa?
Guidance da aka sabunta, wadda Federal Reserve, OCC da FDIC suka fitar a 17 ga Afrilu 2026, ta maye gurbin guidance kan model risk management ta 2011 da abokiyar ta ta 2021. Gine-ginen tushe — ci gaba mai kyau, effective challenge, ongoing monitoring — ya ci gaba.

Shin ƙofar dala biliyan 30 tana nufin ƙananan bankuna za su iya yin watsi da wannan?
A'a. Guidance ɗin ta faɗi cewa ana sa ran za ta fi dacewa da ƙungiyoyi masu girma sama da haka, kuma ta ƙara da cewa tana iya dacewa da ƙananan ƙungiyoyi masu babban fallasa ga model risk saboda yawa da rikitarwar samfuransu, ko ayyuka a wajen bankin al'umma na gargajiya. Bayanin proportionality ne, ba layin keɓancewa ba.

Yaya wannan ke haɗuwa da SS1/23 ta Birtaniya?
Ta rabu da ita. SS1/23, wadda ta fara aiki a 17 ga Mayu 2024, tana tsaka-tsaki kan fasaha kuma ba ta cire AI ko machine learning ba. Bankin ƙungiya da ke gudanar da samfuri guda a hurumi biyu yanzu yana fuskantar amsoshin scope daban-daban, wanda shi ne dalilin da ya sa yin shugabanci bisa ma'auni mafi tsanani shi ne hanyar aiki.

Shin ya kamata mu cire kayan aikin generative AI daga model inventory ɗinmu tunda suna out of scope?
A'a — kuma wannan ita ce shawarar da za a fi yin nadama a kanta. Cikar lissafi tana da arha wajen kiyayewa kuma tana da tsada wajen sake ginawa. A maimakon haka, yi wa waɗannan tsarukan alama a matsayin masu jiran rarrabawa, don rajista ta nuna abin da aka aiwatar da gaske lokacin da tsarin ya zo.

Manazarta

Board of Governors of the Federal Reserve System, 2026. SR 26-2: Revised Guidance on Model Risk Management. Washington, DC: Federal Reserve. Ana samu a: Board of Governors of the Federal Reserve System, 2026..
Office of the Comptroller of the Currency, 2026. OCC Bulletin 2026-13, Model Risk Management: Revised Guidance. Washington, DC: OCC. Ana samu a: Office of the Comptroller of the Currency, 2026..
Prudential Regulation Authority, 2023. SS1/23, Model risk management principles for banks. London: Bank of England. Ana samu a: Prudential Regulation Authority, 2023..
European Parliament and Council of the European Union, 2024. Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Brussels: Official Journal of the European Union. Ana samu a: European Parliament and Council of the European Union, 2024..

Bita ta ƙarshe 2026-08-08.

Sake buga wannan labari

Out of Scope Ba Fita Daga Haɗari Ba: Sabuntawar Model Risk ta 2026 Ta Bar Samfuran da Bankuna Ke Aiwatarwa — Sebastien Rousseau

Sabuntawar model risk management na Afrilu 2026 ta sanya generative da agentic AI a fili out of scope. Fita daga scope ba fita daga haɗari ba — Birtaniya ta ƙi bi.

An lasisin wannan labari a karkashin Creative Commons Attribution 4.0 International. Sake bugawa na bukatar nuna asalin URL na asali.

Out of Scope Ba Fita Daga Haɗari Ba: Sabuntawar Model Risk ta 2026 Ta Bar Samfuran da Bankuna Ke Aiwatarwa — Sebastien Rousseau

Sabuntawar model risk management na Afrilu 2026 ta sanya generative da agentic AI a fili out of scope. Fita daga scope ba fita daga haɗari ba — Birtaniya ta ƙi bi.

Originally published at https://sebastienrousseau.com/ha/2026-07-30-model-risk-management-generative-ai-out-of-scope-2026/ by Sebastien Rousseau.
Licensed under CC-BY-4.0.

SEBASTIEN ROUSSEAU FOUNDER · ENGINEER