Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy

Agentic AI a fannin banki ya tsallaka daga gwaji zuwa kayan aikin gudanarwa. Tambayar da ake yi a 2026 ba ta ko za a tura shi ba ce — kashi 52% na cibiyoyin kuɗi sun riga sun yi — sai dai ko masana'antar za ta iya auna abin da ta gina da irin tsananin da take amfani da shi a jari, bashi, da ruwan kuɗi. Wannan ma'auni shi ne tsarin auna (Cambridge CCAF, 2026).

Taƙaitaccen Bayani ga Manyan Hukumomi / Mahimman Abubuwan da za a Ɗauka

• Cin gashin kai shi ne sabon isasshen jari. Kamar yadda Basel ya kafa mizanai masu aunawa na juriyar kuɗi, yanzu fannin yana buƙatar mizani mai aunawa na yanke shawara mai cin gashin kai. Wannan ma'auni shi ne tsari na farko mai fadin sassa da yake auna shirin agentic AI a fadin shugabanci, tsarin fasaha, hujjar tsari, dawowar tattalin arziki, da girman ƙungiya a matsayin tsarin aiki guda ɗaya.
• Tu`rawar kashi 52% tana ɓoye matakin sauyi na kashi 14%. Binciken Cambridge CCAF na 2026 na ƙungiyoyi 628 a hukunce-hukunce 151 ya gano cewa duk da cewa cibiyoyin kuɗi huɗu cikin biyar suna tura AI, kashi 14% kawai ne ke bayyana shi a matsayin mai sauya matsayinsu na gasa. Gibin shugabanci ne, ba fasaha ba.
• OSWorld a kashi 66.3% shi ne iyakar dogaro, ba bene ba. Ma'aunin Stanford HAI na 2026 ya nuna AI agents suna kammala kashi 66.3% na ayyukan kasuwanci masu tsari (Stanford HAI, 2026). Tool-calls guda uku da aka haɗa a wannan matakin suna haɗuwa zuwa nasarar kashi 29% daga farko zuwa ƙarshe. Aiwatarwa ba tare da kulawa ba a kan tsarin biyan kuɗi mai aiki ba ta da tabbas a wannan matakin dogaro.
• FSB ta yi magana. A 10 Yuni 2026, Financial Stability Board ta buga tsarin gudanarwa na farko don sarrafa agentic AI a fannin kuɗi (FSB, 2026) — sound practices 12 da suka shafi alhakin kwamitin gudanarwa, sarrafa zagayowar rayuwa, da gine-ginen AI-monitoring-AI. Sharhe-sharhe suna rufewa a 22 Yuli 2026.
• Agogon tilasta bin EU AI Act yana tafiya. Wajibcin tsarin AI mai haɗari babba a ƙarƙashin Annex III zai fara aiki a 2 Agusta 2026 (EU AI Act guidance, 2026). Cibiyoyin kuɗi da ke gudanar da agentic AI na EU ba tare da shaidar audit-log ta kowane agent ba, hanyoyin sokewa da aka rubuta, da hujja a matakin kwamitin gudanarwa, suna cikin jinkiri.
• JP Morgan ta ambaci shekara. Derek Waldron, babban jami'in nazari, ya tabbatar wa CNBC a 9 Yuni 2026 cewa bankin zai tura agents masu cin gashin kai masu dogon aiki (CNBC, 2026) — masu iya aiki da kansu na tsawon sa'a ɗaya zuwa biyu — cikin 2026. Wannan bayyanawa ya sauya yanayin gasa ga kowace cibiya da take auna kanta da ita.
• Ma'aunin yana auna sassa shida. Matakin Cin Gashin Kai, Tsarin Shugabanci, Hujjar Tsari, Lissafin Tattalin Arziki, Shirin Ƙungiya, da Daidaiton Tsarin Duniya. Tare suna juya shirin AI daga tarin ayyuka zuwa iyawa mai aunawa.

Me Yasa Wannan Ma'auni Ke Wanzuwa

Evident AI Index yana matsayin bankuna 50 na duniya a fadin Hazaka, Ƙirƙira, Jagoranci, da Gaskiya ta amfani da miliyoyin bayanan da ake samu a fili. Shi ne ma'aunin waje mafi aminci na girman AI a fannin kuɗi. Abin da ba ya yi — da gangan — shi ne auna takamaiman tsarin injiniya da shugabanci da ke sa agentic AI ya zama mai aminci don turawa a kan APIs na banki masu aiki. Stanford AI Index yana bin diddigin sakamakon bincike, aikin fasaha, da tasirin al'umma. Abin da ba ya yi shi ne juya kashi-kashin kammala ayyukan OSWorld zuwa tsarin umarni na gudanarwa ga ma'aji, babban jami'in haɗari, ko ƙungiyar tabbatar da samfuri.

Wannan ma'auni yana cike wannan gibi. Yana ɗaukar tarbiyyar aunawa ta tsarin Stanford, yanayin gasa na Evident Index, da takamaiman tsarin doka na SR 11-7, SS1/23, EU AI Act, sound practices na FSB, da Tsarin Shugabancin AI na Singapore IMDA Model AI Governance Framework for Agentic AI — yana juya su zuwa samfurin auna mai fadin sassa shida da kwamitin gudanarwa zai iya aiki da shi.

Abin da ya jawo wannan a aikace shi ne agentic AI ya sauya daga tattaunawar shiri zuwa tambayar bincike. Lokacin da babban jami'in nazari na JP Morgan ya sanar da turawar agents masu cin gashin kai masu dogon aiki a cikin shekara guda, lokacin da DBS ke gina agent control planes cikin shirya credit memo da hidimar abokan ciniki, lokacin da FSB ke ba da umarni cewa agents da ke aiwatar da hada-hadar kuɗi suna buƙatar "amincewar mutum ko izini biyu sama da darajar iyaka, iyakancewar damar agent zuwa tsarin biyan kuɗi, da audit trails na kowace hada-hadar agent" — cibiyar da ba za ta iya auna matsayinta ba za ta sami mai kula da tsari yana auna ta a madadinta.

Yanayin Girman Agentic AI na 2026

Abin da Bayanai Ke Nunawa

Rahoton Cambridge CCAF na 2026 — babban binciken duniya na AI a fannin kuɗi, da ya shafi ƙungiyoyi 628 a fadin hukunce-hukunce 151 cikin haɗin gwiwa da BIS, IMF, WEF, da Bankin Duniya — yana ba da tushen ƙididdiga na wannan ma'auni.

Waɗannan lambobi suna bayyana matsalar daidai: turawa ya zarce shugabanci, ribar samar da kayayyaki na bayyane, sauyi ba kasafai ba ne, kuma gibin aunawa ya fi fadi inda kaso na tsari ya fi girma — manyan cibiyoyin kuɗi.

Inda Masu Gasa Ke Zana Layi

Evident AI Index 2025 ya sanya JP Morgan Chase na farko (maki: 79), sai Capital One (78.1), RBC (58.4), CommBank Australia (53.9), da Morgan Stanley (52.2). Ma'aunin yana auna ginshiƙai huɗu na iyawa — Hazaka, Ƙirƙira, Jagoranci, Gaskiya — ba tsarin agent na gudanarwa ba. Wannan yana haifar da gibin tsari: banki na iya samun maki mai girma a bayyana Ƙirƙira yayin da yake tura agents marasa kill switch, marasa WORM audit log, da marasa OPA policy gate. An tsara wannan ma'auni don sanya wannan gibi a bayyane.

Rahoton Deloitte na 2026 Tech Trends ya bayyana cewa kashi 11% kawai na ƙungiyoyi ne ke da agentic AI cikin gudanarwa. McKinsey ya gano cewa kusan ɗaya bisa uku na ƙungiyoyi kawai ke kaiwa matakin girman shugabanci na uku ko sama da haka cikin sarrafa agentic AI ko da yake iyawar fasaha tana ci gaba cikin sauri. Bayanan binciken CCG Catalyst sun nuna kashi 93% na kuɗaɗen da ke da alaƙa da AI suna zuwa kayan fasaha kuma kashi 7% kawai zuwa mutane, hazaka, horarwa, sarrafa sauyi, da shugabanci — rabo da ke sa fadadawa ya zama ba mai yiwuwa ba a tsarin.

Evident Venture Tracker na Q1 2026 ya gano Anthropic a matsayin mai siyarwa da aka fi ambata, tare da dogon jeri na 'yan wasa na musamman da ke da kashi 68% na duk turawa, galibi suna nufin amfani na musamman na aiki cikin bashi, yaƙi da kuɗin haram, da baitulmali. Bangaren samarwa ya balaga. Bangaren shugabanci bai balaga ba.

Tsarin Ma'auni Mai Sassa Shida

Wannan ma'auni yana auna shirin agentic AI a fadin sassa shida. Kowane sashe yana da ma'aunin girma mai matakai huɗu. Makin ma'aunin banki shi ne sakamakon makin sassansa da aka auna da kaso na tsari. An daidaita tsarin auna nauyi da SR 11-7, SS1/23, wajibcin EU AI Act Annex III, da nau'ikan Sound Practice na FSB.

Sashe na 1: Rufin Matakin Cin Gashin Kai

Abin da yake aunawa: Ko an rarraba kowane workflow na agentic na gudanarwa a kan tsani na cin gashin kai da aka ayyana, ba tare da wani workflow na aiki sama da matakinsa da aka ba da izini ba tare da banbanci da aka rubuta — kuma ko wannan ayyana matakin yana bayyana ba kawai iyakar aiki ba har ma da iyakar alhakin doka.

Tsani na cin gashin kai ya kasance ginshiƙin tushe. Matakai biyar — daga Level 0 (lura da karatu kawai) zuwa Level 4 (haɗa kayan aiki da yawa tare da matakan duba na tilas) — suna bayyana iyakar izini na agent, ba ƙwarewar samfuri ba. Wannan LLM ɗaya na iya zama a kowane mataki; wrapper shi ne abin da ya bambanta. Level 5 — aiwatarwa mai kula da kanta ba tare da matakan duba ba — bai kamata ya wanzu cikin gudanar da banki a 2026 ba. OSWorld a kashi 66.3% na kammala aiki yana haɗuwa: tool-calls guda uku da aka haɗa kowannensu a kashi 66% yana samar da nasarar kashi 29% daga farko zuwa ƙarshe. Haɗi biyar yana samar da kashi 13%.

Tsarin Shugabancin AI na Singapore IMDA Model AI Governance Framework for Agentic AI, da aka buga a Davos a 22 Janairu 2026 a matsayin tsarin shugabanci na farko a duniya da ya magance agents masu cin gashin kai a sarari (IMDA, 2026), yana bayyana ra'ayoyi huɗu masu daidaito: matsayin shugaba (wa zai iya umartar agent), iyakar aiki (abin da aka ba agent izinin yi), ƙaramin sawu (agent bai kamata ya tara izini fiye da buƙatar gaggawa ba), da iya bayyanawa (dole hanyoyin tunani su zama masu iya binciko). Waɗannan huɗu suna haɗuwa kai tsaye da samfurin matakin cin gashin kai.

Matsalar Shugaba-Wakili da Danganta Niyya ta Doka. Tsarin IMDA yana gabatar da wani sashe da takamaiman ƙayyadaddun injiniya ke raina: lokacin da AI agent ke aiki a matsayin wakilin wata kamfani — yana aiwatar da biyan kuɗi, yana amincewa da gyaran iyakar bashi, yana gabatar da takardar tsari — yana haifar da matsalar danganta niyya ta doka. A ƙarƙashin ikon wa agent ya yi aiki? Wa zai ɗauki alhaki lokacin da agent ya saɓa wa ƙuntatawar prompt ɗinsa? Niyyar wa ake dangantawa lokacin da agent ya zaɓa tsakanin fassara biyu masu inganci-amma-daban na umarni mai shubuha?

Ga workflow na Level 3 da Level 4 — inda agent ke aiwatar da ayyuka masu sakamako da kansa cikin sigogi da aka ayyana — ayyana matakin dole ya bayyana ba kawai iyakar aikin fasaha ba har ma da iyakar alhakin doka: shugaba mutum mai suna da ya ba da izinin workflow, kayan aikin wakilci da aka rubuta (ƙudurin kwamiti, wakilcin iko, ko izini da aka sa hannu), yanayin da ayyukan agent ke ɗaure cibiyar, da yanayin da saɓawar ƙuntatawar prompt ke jawo sokewa ta atomatik, ƙararrawa, da rubuta abin da ya faru. Idan ba haka ba, rarrabuwar matakin cin gashin kai abu ne na injiniya da ba zai tsira daga ƙalubalen doka ba, bincike na tsari, ko jayayya da abokin huɗu wanda kuɗinsa ya motsa saboda agent ya yi kuskuren fassara umarni mai sharaɗi.

Sashe na 2: Tsarin Shugabanci

Abin da yake aunawa: Ko agent control plane mai sassa biyar an gina shi gaba ɗaya kuma yana aiki cikin gudanarwa — ba wai an bayyana shi a takardar manufa ba.

Tuntubar FSB na Yuni 2026 ta bayyana a sarari cewa tsarin shugabanci da ke akwai ba a tsara su don tsarin da ke "shiryawa, ɗaukar matakai da yawa, da hulɗa da tsarin waje ba tare da kulawar mutum mataki-mataki ba". Control plane mai sassa biyar yana juya wannan lura zuwa jerin abubuwan dubawa na injiniya:

Sashe na 1: Tantancewa da Izini. Kowane agent yana danganta zuwa daidai service account guda ɗaya tare da OAuth client_credentials tokens da aka iyakance zuwa mafi ƙarancin API surface. Token na agent na daskare-katin na iya kira POST /accounts/{id}/freeze tare da iyakar adadi; ba zai iya kira komai a riƙo, baitulmali, ko ciniki ba. Sirrikan service-account suna juyawa a kan zagayowar da aka ayyana. Shaidu masu dadewa su ne gazawar control-plane da ta fi yawa cikin turawar gudanarwa. FSB ta ba da shawara a sarari "mafi ƙarancin gata ga agents da agents ɗinsu na ƙasa, da sarrafa tantancewa da izini mai motsi da ke bayarwa, sauyawa ko soke izini cikin lokaci na ainihi bisa hali da yanayi, maimakon bayyanannun bayanai masu tsayawa da ake amfani da su ga masu amfani na mutane".

Sashe na 2: Guardrails Masu Tabbas. Kowace LLM tool-call tana wucewa ta semantic router (NeMo Guardrails, LangChain Guardrails, ko makamancinsu) kafin ta kai ga API na gudanarwa. Router yana rarraba niyya a kan allow-list mai iyaka kuma yana ƙin kira a wajen wannan jeri. Sai JSON-schema validator ya duba payload. pacs.008 mai amount: 0 gazawar samfuri ce, ba hada-hadar halal ba. Haka kuma wire zuwa ƙasar da ba a riga an amince da ita ba ga sashin abokin ciniki na asali.

Sashe na 3: Policy-as-Code. Open Policy Agent (ko makamancinsa) yana tsakanin validator da API. Ana sarrafa nau'ikan manufofin cikin Git; ana rubuta shawarwarin ƙi; injin manufa ɗaya da ke kula da kira tsakanin microservice cikin tsarin da ke akwai yana kula da agent tool-calls. Jagorar EU AI Office na Mayu 2026 kan rubuta audit na Article 12 tana buƙatar cewa shigarwar log na tsarin AI mai haɗari babba su danganta ayyuka zuwa takamaiman agent instance, ba kawai turawa ko shaidar API ba. Turawar agent da yawa da ke raba shaida guda suna gaza wannan gwaji.

Sashe na 4: Cikar Bincike. Ajiyar WORM da ba a iya canzawa — S3 Object Lock, Azure Blob immutability, ko bayanan da aka rubuta a littafi. Kowane kira yana ɗaukar: lokaci, agent ID, service-account ID, hash na system-prompt, yanayin da aka dawo da shi, mai bayar da LLM da samfuri da sigar, ɗanyen fitar LLM, parsed tool-call, shawarar OPA, amsar API, tasiri na ƙasa, da approver UID inda ya dace. Ana sa hannu na cryptographic kan bayanan a lokacin rubutawa. Bayanin Article 12 na EU AI Act da aka buga Mayu 2026 ya ambaci tantancewar kowane agent a matsayin takamaiman gibi; cibiyoyi da ke gudanar da agent instances da yawa da ke raba shaida guda a sarari sun karya tsari.

Sashe na 5: Kill Switch da AI-Monitoring-AI. API na maɓallin ja da aka gwada da ke soke duk kiraye-kirayen agent masu tafiya cikin ajin izini cikin ƙasa da daƙiƙa 60. Kalmar gwada na ɗauke da nauyi. Kill switch da ba a gwada ba burin manufa ne.

Bayan kill switch, Sashe na 2 a mataki mafi girma na girma dole ya tilasta tsarin AI-monitoring-AI (AMI) — kuma dalilin lissafi ne. Bayanan IBM sun sanya matsakaicin yawan agent na kasuwanci a 1,661 nan da 2027 (IBM, 2026). FSB ta yarda a sarari cewa ci gaba da kulawar mutum kan shawarwarin agent ɗaiɗai yana zama ba mai yiwuwa ba a girma, kuma tana ba da shawarar ƙara wa kulawar mutum tsarin AI da ke faɗakar da mutane lokacin da aka karya ma'aunin aiki ko halin agent ya ɓace. Jami'in bin doka na mutum ba zai iya sa ido kan agents 1,661 a lokaci guda da ke aiwatar da shawarwari cikin saurin na'ura ba. Tsarin sarrafawa da ke ɗauka cewa za su iya zai gaza a karon farko da yawan agent ya sami sauyin hali mai alaƙa — sabunta samfuri da ke canza rabon fitarwa a shiru a fadin workflow da yawa a lokaci guda.

Layin AMI ba maye gurbin kulawar mutum ba ne; shi ne hanyar ganowa da ke sa kulawar mutum ta zama mai aiki a girma. Ayyukansa uku na tilas su ne: ganowar ɓacewa (sa ido na ƙididdiga kan rabon fitarwa a fadin agents na matakin da nau'in iri ɗaya, yana yiwa saɓawar da ta zarce ma'aunin sigma da aka ayyana alama kafin mutum ya iya lura da ita); ƙararrawar alaƙa tsakanin agents (gano lokacin da agents da yawa suka fara aiwatarwa cikin tsari mai daidaito wanda ba ya nan jiya — alamar farko ta yanayin tafiya tare da aka bayyana a Sashe na 6); da gargaɗin gaba kan abin da bai dace ba (samar da ƙararrawa mai tsari, tare da yanayi da tantance iya juyowa, ga mai yanke shawara mutum kafin kill switch ya zama zaɓi guda da ya rage). FSB tana ba da shawarar gine-ginen AMI a sarari a Sound Practice 9. Cibiyar da ta kai Matakin Girma 4 a Sashe na 2 ba tare da layin AMI mai aiki ba ba ta kan Level 4 ba.

Sashe na 3: Cikar Hujjar Tsari

Abin da yake aunawa: Ko cibiyar na iya samar da cikakken kunshin hujjar tsari ga kowane workflow lokacin da ake buƙata ga SR 11-7, SS1/23, EU AI Act, DORA, FSB, da tsarin ƙasa masu dacewa.

Federal Reserve ta sake bayyana kai-da-kai cewa SR 11-7 ya shafi kowane tsarin yanke shawara daga shigarwa zuwa fitarwa, ko da kuwa cibiyar ta rarraba LLM na asali a matsayin samfuri ko a'a. SS1/23 na PRA ya fi fadi. Rarrabuwar haɗari babba na Annex III na EU AI Act ya shafi yawancin amfani na LLM na fannin kuɗi — ƙididdigar bashi, gano zamba, dacewar abokin ciniki, farashin inshora. Ana buƙatar cikakken bin doka ga tsarin EU nan da 2 Agusta 2026, tare da Jamus, Faransa, da Netherlands da aka tabbatar don dubawar kulawa na Q3 2026. IOSCO Supervisory Toolkit for AI Use in Capital Markets, da aka kammala a 25 Mayu 2026, ya shafi cikakkiyar zagayowar rayuwar AI daga ML na gargajiya ta GenAI da agentic AI — kuma yana gano a sarari cewa iyawar shiryawa, ƙwaƙwalwa mai dadewa, da damar kayan aiki na waje suna haifar da haɗarin halin da ke fitowa da gazawa masu yaduwa a fadin tsarin da ke da alaƙa.

Samfurin layi-uku-na-tsaro, da aka yi amfani da shi ga agents:

• Layi na farko (mai mallakar samfuri): Yana rubuta amfanin da aka nufa, asalin bayanan horarwa da tantancewa, schema na system-prompt, allow-list na tool-call, sakamakon gwajin kill-switch. Yana mallakar sa ido kan ɓacewa cikin gudanarwa. Yana mallakar saitin tantancewa na musamman na banki da aka ware — aikin da yawancin cibiyoyi ke ƙarancin zuba jari a ciki.
• Layi na biyu (ƙungiyar MRM): Tana tabbatar da agent kafin gudanarwa. Rahoton tabbatarwa ya shafi makin tantancewa na mai siyarwa (MMLU, HumanEval — masu amfani amma ba isasshe ba), makin tantancewa na musamman na banki, sakamakon red-team na prompt-injection, nazarin nuna bambanci da adalci, da bayanin ragowar haɗari da aka ƙididdige.
• Layi na uku (binciken cikin gida): Yana gwada control-plane gates da cikar audit-log a kan samfurin shawarwarin gudanarwa. Zagayowar binciken 2027 za ta yi kama da ta 2025 sosai daban; ku yi kasafin kuɗi daidai.

Singapore Model AI Governance Framework for Agentic AI (MGF) yana buƙatar cibiyoyin kuɗi su tantance agents a fadin sassa huɗu: iyakance cin gashin kai da damar agent, kafa alhakin mutum a matakan duba da aka ayyana, aiwatar da kayan sarrafa fasaha da suka haɗa da gwajin tushe, da bayar da damar alhakin mai amfani na ƙarshe ta hanyar gaskiya. AI Risk Management Toolkit na MAS na Maris 2026 — da aka haɓaka a ƙarƙashin Project MindForge tare da cibiyoyi 24 — yana wakiltar jagorar matakin ƙasa mafi cikakke a aikace da ake samu.

Sashe na 4: Lissafin Tattalin Arziki

Abin da yake aunawa: Ko cibiyar tana auna dawowar agentic AI ta amfani da tattalin arziki na naúrar matakin workflow maimakon da'awar ribar matakin shirin.

Nazarin McKinsey ya gano cewa agentic AI na iya rage kuɗaɗen gudanar da banki da kashi 15–20% (McKinsey, 2026) — daidai da kashi 9–15% na ribar gudanarwa — amma yawancin waɗannan ribobin za a yi gasa a kansu. Fa'idar gasa mafi dorewa tana cikin cibiyoyin da suka gina tsarin auna don yin aiki cikin sauri fiye da masu gasa lokacin da haɓakar samfuri da workflow suka samu. Binciken Cambridge CCAF cewa kashi 76% na manyan cibiyoyin kuɗi ba sa iya auna darajar turawar AI ba matsalar ingancin bayanai ba ne. Matsalar tsarin-alhaki ce: ana yin kasafin kuɗi da ba da rahoton shirye-shirye a matakin fayil, yana sa ba zai yiwu a binciko daraja ko gazawa zuwa workflow ɗaiɗai ba.

Ma'aunin tattalin arziki na naúra huɗu da ke tsira daga tattaunawar CFO:

Kuɗin kowace shawarar da aka kammala, gami da kuɗin sokewa da gyara na shawarwarin da suka gaza. Agent na shirya SAR da ke rage lokacin jami'in BSA da kashi 40% amma yana samar da kashi 12% na fayil na ƙarya ya lalata daraja, ba ya halicce ta ba. Wannan shi ne ma'aunin da binciken Deloitte — cewa kashi 93% na kuɗaɗen AI suna zuwa kayan aiki kuma kashi 7% kawai zuwa mutane da shugabanci — ya sa ba mai aunawa ba: cibiyoyi ba sa iya lissafa kuɗin sokewa na gazawar shugabanci da ba su shirya don ganowa ba.

Taɓawar hannu da aka kauce, da aka ƙidaya bayan cire sabbin taɓawa da aka samar ta kulawar control-plane da sarrafa banbanci. Manufar ba rage hankalin mutum ba ce; ita ce karkata da shi zuwa shawarwari masu fa'ida mafi girma.

Adadin sokewa — kashi-kashin ayyukan da agent ya aiwatar da aka mayar baya cikin sa'o'i 24. Workflow na Level-3 mai adadin sokewa sama da kashi 2% matsalar dogaro ce. Sama da kashi 5% matsalar control-plane ce. Ya kamata a bi diddigin wannan lambar ga kowane workflow, ba kowane shiri ba. Matsakaicin fayil yana ɓoye wanda ya fita waje da zai samar da binciken da zai biyo baya.

Cikar audit-trace — kashi-kashin shawarwari masu cikakkiyar tushen da ake iya sake ginawa daga WORM log. Ya kamata ya zama kashi 100% kan workflow na Level-3 da Level-4. Komai ƙasa da haka gazawar manufa ce.

Kasuwar agentic AI a fannin banki tana girma a wani matakin da ke sa wannan tsarin auna ya zama gaggawa. Rahoton Newgen na 2026 Banking Trends ya hango kasuwar agentic AI tana girma daga dala biliyan 2.1 zuwa dala biliyan 81 nan da 2034. Tsarin yanayi na McKinsey ya nuna cewa sakamakon da ya fi yiwuwa — yanayin yiwuwa na kashi 30% — ya shafi AI agents da ke kaiwa rabon agent-zuwa-mutum kusan 20:1 da samar da raguwar kuɗi na kashi 15–20%. Majagaba na iya buɗe gibin maki kashi 4 na ROTE dangane da masu jinkiri. Wannan gibin na gaske ne, amma ana iya auna shi da kare shi ne kawai idan ana bin diddigin tattalin arzikin naúra a matakin workflow.

Sashe na 5: Shirin Ƙungiya

Abin da yake aunawa: Ko cibiyar tana da hazaka, shugabanci na fannoni daban-daban, rahoton matakin kwamiti, da al'ada don turawa da ci gaba da agentic AI a girma — ba kawai gwada shi ba.

Binciken Cambridge CCAF daidai ne: shirin ma'aikata yana hangen ribar AI sau huɗu fiye da sayan fasaha. Kamfanonin da ma'aikata suka shirya sosai suna ba da rahoton ribar AI kashi 23%; kamfanonin da ba su shirya ba suna ba da rahoton kashi 6%. Kashi 10% kawai na duk kamfanoni ke bayyana ma'aikatansu a shirye. Fintechs suna kaiwa matakin sauyi sau uku fiye da cibiyoyin kuɗi na gargajiya — kashi 19% da kashi 6% — duk da yawancinsu suna kashe ƙasa da dala 10,000 a shekara kan AI. Tsarin shi ne mai bambantawa, ba kasafin kuɗi ba.

McKinsey yana bayyana matsayoyi uku na dabaru ga bankuna da ke fuskantar agentic AI: jira da gani, daidaitawa ta hanyar zama mai samar da kayayyaki a bayan agent interfaces, ko gasa don mallakar dangantakar abokin ciniki kai tsaye. Yawancin bankuna sun koma matsayi na farko yayin da suke wakilta kansu suna bin na uku. Dole tattaunawar dabaru ta zama a sarari, kuma kwamiti shi ne inda dole ta sauka.

FSB Sound Practice 1 yana magana kai tsaye kan alhakin kwamiti: kwamitoci suna ɗaukar alhaki na ƙarshe na shugabancin AI, kafa sha'awar haɗari, da tabbatar da cewa tsarin alhaki na bayyane ne. Tilasta EU AI Act Article 5 da tanade-tanaden alhakin kwamiti na DORA Article 5 suna juya wannan ƙa'ida zuwa alhakin mutum. Supervisory Toolkit na IOSCO na Mayu 2026 ya bayyana cewa "tsarin AI ba ƙananan ayyukan da aka ware ba ne. Su ne tushen kayan aikin gudanarwa da ke buƙatar ci gaba da tabbatarwa, shugabanci na matakin kwamiti, da hujjar kulawa a shirye don bincike".

Tsarin rahoton kwamiti ga agentic AI ya kamata ya shafi lambobi huɗu ga kowane workflow: matakin cin gashin kai, cikar audit-trace, adadin sokewa, da kuɗin shawara na yanzu. Tare da jerin manyan biyar na ragowar haɗari. Slideware na takardar manufa ba maye gurbi ba ne.

Sashe na 6: Daidaiton Tsarin Duniya

Abin da yake aunawa: Ko tsarin aiki na agentic AI na cibiya ya yi daidai da manyan tsarin tsari huɗu da ke aiki a manyan hukunce-hukuncen aikinta — kuma ko an tabbatar da wannan daidaito da hujja, ba a faɗi shi kawai ba.

Yanayin tsari na agentic AI ya tabbata a rabi na farko na 2026. Tsarin huɗu yanzu suna da mahimmancin aiki:

Amurka (SR 11-7 / OCC Bulletin 2025-26). Jagorar sarrafa haɗarin samfuri na Federal Reserve tana aiki ga kowane workflow na yanke shawara na LLM. OCC ta buga takamaiman jagorar sarrafa haɗarin samfuri ga bankunan al'umma da ke jaddada daidaitawa — "daidaitawa ba ya nufin rashi". Samfurin layi-uku-na-tsaro yana aiki gaba ɗaya.

Birtaniya (PRA SS1/23 / FCA). Ƙa'idodin sarrafa-haɗarin-samfuri na SS1/23 na PRA sun yi fadi sosai don ɗaukar duk agents na LLM. Hukumar kulawa ta Birtaniya tana haɓaka takamaiman tsammanin agentic AI. FCA na cikin hukumomin ƙasa da ke ba da ƙarin jagora kan shugabancin AI a fannin kuɗi.

Tarayyar Turai (EU AI Act / DORA). Wajibcin tsarin AI mai haɗari babba na Annex III na aiki daga 2 Agusta 2026. Buƙatu sun haɗa da sarrafa haɗari mai tsari (Article 9), shugabancin bayanai (Article 10), gaskiya (Article 13), kulawar mutum (Article 14), da rubuta audit na kowane agent (Article 12). Tanade-tanaden alhakin kwamiti na DORA Article 5 sun shafi juriyar aiki gami da agentic AI. Jagorar EU AI Office na Mayu 2026 tana tilasta tantancewar cryptographic ta kowane agent a cikin audit logs. Rashin bin doka yana ɗauke da tara har zuwa EUR miliyan 35 ko kashi 7% na jujjuyawar duniya.

Asiya-Pacific (MAS / IMDA / masu kula da yanki). Singapore IMDA ta buga Model AI Governance Framework for Agentic AI na farko a duniya a Davos a 22 Janairu 2026. MAS ta buga AI Risk Management Toolkit a Maris 2026 a ƙarƙashin Project MindForge, da aka haɓaka tare da cibiyoyin kuɗi 24. Tsarin ya shafi iyaka da kulawar AI, sarrafa haɗarin AI, sarrafa zagayowar rayuwar AI, da masu ba da damar ƙungiya. Ana sa ran kammala Guidelines on AI Risk Management na hukuma da MAS ta gabatar a 2026, suna matsawa daga ƙa'idodin FEAT na son rai zuwa tsammanin kulawa tare da illolin bin doka. ASIC na Australia ya ba da wasiƙar buɗe ido a Mayu 2026 yana neman haɓaka tsaron yanar gizo dangane da barazanar frontier AI.

FSB (Duniya, fadin hukunce-hukunce). Tuntubar FSB na Yuni 2026 — tsari na farko na duniya da ya ɗauki agentic AI a matsayin mai bambanci a aikace — ya gano samfuran kulawa shida ga tsarin agentic kuma yana ba da shawarar mutum-cikin-umarni ga workflow na cin gashin kai babba, sa ido na AI-cikin-zagaye yayin da yawan agent ke girma, da amincewar mutum ko izini biyu ga agents da ke aiwatar da hada-hadar kuɗi sama da darajar iyaka. Sharhe-sharhe suna rufewa a 22 Yuli 2026; rahoton ƙarshe ga ministocin kuɗi na G20 a Oktoba 2026.

Makin Ma'aunin Haɗaɗɗe

Makin sassa shida suna haɗuwa zuwa ma'auni haɗaɗɗe ta amfani da auna nauyin-mahimmancin-tsari mai zuwa:

Makin haɗaɗɗe ƙasa da 50 yana nufin cibiyar ba za ta iya kare matsayinta na agentic AI na yanzu ga mai binciken SR 11-7, dubawar PRA na wuri, ko tantancewar kulawa na EU AI Act ba. Maki na 50–74 yana nufin kayan sarrafawa suna nan amma ba su zama masu ci gaba ko a shirye da hujja ba tukuna. Maki na 75–100 yana nufin shugabanci kadara ce ta gasa, ba kuɗin bin doka ba.

Alamomin Yanzu da za a Bi Diddigi

Abin da Wannan Ke Nufi bisa Nau'in Cibiya

Bankuna Masu Muhimmanci ga Tsarin Duniya (G-SIBs)

G-SIBs suna fuskantar ƙalubalen shugabanci mafi wuya — ba don fasahar ta fi rikitarwa ba, sai dai don girma da hukunci suna haɗa kowane gibi. G-SIB mai agents 200 na gudanarwa a fadin layukan kasuwanci 30 a hukunce-hukunce 15 na tsari yana da binciken SR 11-7 200 masu yiwuwa, gazawar audit-log na EU AI Act 200 masu yiwuwa, da gibin FSB Sound Practice 200 masu yiwuwa — a lokaci guda. Fifikon zuba jari ba wani gwaji ba ne. Shi ne control plane na tsakiya, kayan aikin audit-log haɗaɗɗe, da benci na MRM mai iya tabbatar da agents 50-da-sama kowane kwata.

Sanarwar JP Morgan na agents masu cin gashin kai masu dogon aiki a 2026 — agent control planes na DBS cikin shirya credit memo da hidimar abokan ciniki — BNP Paribas da ke cika manufofin AI na 2025 da fara rahoton ROI na kowane kwata — waɗannan su ne bayanan gasa da kowane kwamitin G-SIB ya kamata ya auna kansa da su. Tambayar cibiya ba ta ko za a tura ba; ita ce ko control plane na iya fadada a daidai matakin yawan agent.

FSB tana gargaɗi a sarari kan haɗarin tarin daga dogaro kan 'yan kaɗan na cloud, hardware, da masu bayar da foundation-model — kuma tana lura cewa samfura da bayanai da aka raba na iya tura cibiyoyi zuwa hali mai alaƙa da ke ƙarfafa tafiya tare da procyclicality cikin koma baya. G-SIBs da ke samun kashi 80% na kayan aikin agentic ɗinsu daga masu siyar da foundation-model guda biyu suna gina alaƙar tsari da za su buƙaci bayyana wa ƙungiyoyin haɗarinsu da masu kula da su.

Tafiya Tare Mai Tsari da Procyclicality: Haɗarin Tsarin da Babu Banki Guda da zai iya Magancewa Shi Kaɗai. Mai bin diddigin amfani na Evident Insights Q1 2026 ya gano cewa kashi 68% na turawar agentic na banki yanzu suna amfani da dogon jeri na masu siyar na musamman — yawancinsu an gina su a kan frontier models na asali iri ɗaya, galibi Claude na Anthropic. Wannan yana haifar da rauni na tafiya tare mai tsari da ya bambanta sosai da haɗarin tarin da bankuna suka riga suna sarrafawa cikin kayan aikin cloud ko hanyoyin biyan kuɗi.

Tsarin yana kamar haka. Agent na ciniki, agent na ruwan kuɗi, da agent na ƙuntata bashi na banki an gina su a kan dandalin masu siyar daban-daban. Suna da system prompts daban-daban, schemas na tool-call daban-daban, OPA policy gates daban-daban. Amma suna raba samfuri na asali iri ɗaya — nauyi iri ɗaya, rabon horarwa iri ɗaya, tsarin halin da ke fitowa iri ɗaya a ƙarƙashin matsin rabuwa. Lokacin da babban abin da ya faru na kasuwa ya faru — abin da ya faru na bashin ƙasa, sadarwar Fed da ta bambanta da yarjejeniya, gazawar babban banki — kowane agent da aka gina a kan samfuri na asali iri ɗaya zai sarrafa abin da ya faru ta auna fasalin a ɓoye iri ɗaya. Idan waɗannan aunawar suka samar da nuna bambanci na hali na nesa da haɗari, agents na ciniki, ruwan kuɗi, da bashi na bankuna da yawa na iya aiwatar da sayar da haɗaɗɗiya, zagayowar ƙuntata bashi, ko cire ruwan kuɗi a lokaci guda — ba don agent na kowane banki ɗaiɗai yana lalacewa ba, sai dai don duka suna aiki daidai a kan samfuri iri ɗaya.

IOSCO ta ambaci wannan yanayi a sarari a Supervisory Toolkit na Mayu 2026, tana gargaɗi cewa iyawar shiryawa, ƙwaƙwalwa mai dadewa, da damar kayan aiki na waje suna haifar da haɗarin halin da ke fitowa da gazawa masu yaduwa a fadin tsarin da ke da alaƙa. Tuntubar FSB na Yuni 2026 ta magance procyclicality kai tsaye — tana lura cewa idan AI agents an horar da su a kan bayanai iri ɗaya kuma suna amfani da samfura masu kama, halinsu na iya zama mai alaƙa, mai yiwuwa yana ƙarfafa motsin kasuwa.

Auna juriyar tafiya tare mai tsari a Sashe na 6 yana buƙatar bayyanawa uku da iko ɗaya na tsari. Bayyanawar: menene foundation model na asali ga kowane workflow na agentic na gudanarwa; menene taswirar dogaro kan mai siyarwa a fadin fayil na agent; da menene tantancewar cibiya kan gudunmawarta ga hali mai alaƙa tsakanin cibiyoyi a ƙarƙashin yanayin matsi da aka ayyana. Iko na tsari: aƙalla ɗaya daga cikin manyan agents a fannonin kadara masu haɗari babba (ciniki, sarrafa ruwan kuɗi, bashi) dole ya yi amfani da samfuri na asali daban ko bambancin fine-tuned mai bambanci sosai, ta yadda martanin rabuwar samfuri guda ga abin da ya faru na matsi ba zai iya samar da sakamako mai cikakkiyar alaƙa a fadin dukkan workflow na agentic a lokaci guda ba. Wannan bambancin samfuri ne a matsayin sarrafa haɗarin-tsari — daidai na agentic na bambance-bambancen abokin huɗu.

Bankunan Hada-hada da Bankunan Kamfani

Workflow na agentic masu ROI mafi girma su ne gyaran biyan kuɗi, fitar takardun KYC, hidimomin baitulmali, karaya na daidaitawa, da kaucewar FAQ na abokin ciniki na kamfani. Duka Level-2 ko Level-3 mai iyaka a ƙarƙashin tsani na cin gashin kai. Abokin ciniki na kamfani bai damu cewa agent ya aiwatar da gyaran biyan kuɗi ba; suna damuwa cewa SLA ya inganta kuma adadin jayayya ya kasance daidai. Ka jagoranci da ma'aunin tattalin arziki na naúra huɗu, ba da'awar iyawar fasaha ba.

Tsarin Autonomous Treasury — lura → ganowa → hango → shirya → neman amincewar mutum → gabatar da payload da aka sa hannu — shi ne tsarin da ya dace ga agents na baitulmali na kamfani a 2026. pain.001 payload da agent ya shirya yana wucewa ta tantance schema, ƙididdigar zamba, da injunan takunkumi iri ɗaya da gabatarwar ERP na kamfani. Layin sharaɗi (iyaka, cancantar tabbaci, bene na buffer) yana kula da ko an aika pain.001, ba siffar da yake ɗauka ba. Dandalin baitulmali da ke ƙirƙira payloads na musamman don bayyana sharuɗɗa za su fadi daga hanyar da banki ke iya cinyewa.

Bankunan Yanki da Bankunan Al'umma

Nazarin yanayi na McKinsey ya gano matsayoyi uku masu yiwuwa: jira da gani, daidaitawa a matsayin mai samar da kayayyaki a bayan agent interfaces, ko gasa don dangantakar abokin ciniki kai tsaye. Bankunan yanki da suka kasa yin wannan zaɓi a sarari za su koma matsayin jira-da-gani ta tsohuwa — kuma su gano cewa bashin shugabanci da aka tara a lokacin wannan koma shi ne babban cikas lokacin da matsin gasa ya tilasta aiki.

Ƙa'idar daidaitawa ta OCC — "daidaitawa ba ya nufin rashi" — ita ce tsarin aiki ga shugabancin yanki. Banki na yanki ba ya buƙatar tabbatar da agents 50 kowane kwata. Yana buƙatar jami'in haɗarin samfuri guda da ya fahimci tsani na cin gashin kai, aiwatarwa guda na dandalin agent na mai siyarwa da ke zuwa da OAuth scoping, haɗin OPA, da WORM audit logging a shirye, da samfurin rahoton kwamiti guda da ya shafi ma'aunin tattalin arziki na naúra huɗu. Zuba jarin yana cikin ƙirar workflow da UX na mai aiki, ba injiniyan control-plane na musamman ba.

Binciken CSI na 2026 Banking Priorities ya gano cewa kashi 85% na masu amsa banki na al'umma sun yi imani turawar AI za ta ba da fa'idar gasa mai muhimmanci kuma kashi 50% sun ambace shi a matsayin babban yanayin fasaha na 2026. Kayan aikin shugabanci shi ne abin da ke raba kashi 85% na masu imani da ƙaramin sashe da zai kama darajar.

Fintechs, PSPs, da Masu Bayar da Kayan Aiki

Tambayar kayayyaki ga masu siyar da agentic AI a 2026 ba ta "shin dandalinka yana aiki fiye da mutane?" Ita ce "shin dandalinka yana samar da audit trace mai bin SR 11-7, log na kowane agent mai bin EU AI Act Article 12, da samfurin kulawa mai bin FSB Sound Practice 10 — a shirye?" Masu siyar da za su iya amsa hakan da eh da aka rubuta kuma mai gwadawa za su rufe yarjejeniyoyin kasuwanci. Masu siyar da ba za su iya ba za su yi zagaye ta hanyoyin proof-of-concept yayin da ƙungiyoyin MRM na banki ke neman dalilan gaza tabbatarwa.

Oracle ta ƙaddamar da dandalin agentic AI na kasuwanci ga banki a Fabrairu 2026. FIS ta haɗa hannu da Mastercard da Visa don ba da damar kasuwancin da agent ya fara. Microsoft ta buga tsari na musamman na banki ga ƙwarewar abokin ciniki na agentic. Accenture ta zayyana illolin ma'aikata a fadin ofishin gaba da na baya. Bangaren samarwa a shirye yake. Bambanci yana cikin hujjar tsari a matsayin fasalin kayayyaki, ba ƙarin bin doka na bayan-fage ba.

Yanayin mai siyar na dogon jeri da Evident ya gano — kashi 68% na turawar agentic AI a bankuna yanzu suna amfani da masu siyar na musamman fiye da hyperscalers — yana nufin haɗarin mai siyar da AI na huɗu yana hanzari da sauri fiye da yadda yawancin tsarin sayan banki ke iya tantance shi. DORA tana buƙatar tabbacin da aka rubuta kan kowane mai bayar da ICT na huɗu. EU AI Act tana ƙara buƙatu ga masu siyar da ake amfani da tsarinsu a nau'ikan haɗari babba. Bankunan da ke ba da shugabanci ga mai siyar ɗinsu suna ba da alhaki — kuma rikodin kulawa zai nuna hakan.

Kamfanonin Kasuwanci da SME (Hidimomin Kuɗi na Wajen Banki)

Nauyin shugabanci yana daidaita da mahimmancin haɗarin amfani na agentic AI, amma tsarin auna yana aiki a ko'ina. Kamfanin da ke tura agents cikin biyan kuɗi, haɓaka jarin aiki, ko shiri da nazarin kuɗi yana buƙatar tsarin lissafin tattalin arziki na naúra iri ɗaya — kuɗin kowace shawarar da aka kammala, adadin sokewa, cikar audit-trace — ko da wajibcin tsari ya fi sauƙi fiye da na banki mai muhimmanci ga tsarin duniya. An tsara FSB Sound Practices a matsayin jagora mara ɗaure da ta dace da cibiyoyin kuɗi na kowane nau'i da girma. Binciken IBM cewa kamfanoni suna da matsakaicin abubuwan AI agent 54 a shekara, gami da ɓarkewar bayanai da gazawar tsari masu yaduwa, yana aiki a fadin yanayin kasuwanci.

Ga SME masu samun hidimomin banki ta hanyar agent interfaces — yanayin da McKinsey ya bayyana a matsayin masu siye suna amfani da AI agents a matsayin sabuwar hanyar banki — wajibcin shugabanci yana faɗuwa sama kan banki ko PSP da ke bayar da layin agentic. Amma bayanan SME da amincin aikinta sun dogara da wannan shugabanci ya zama na gaske. Fahimtar makin ma'aunin cibiyoyin da ke sarrafa workflow ɗinka na kuɗi yana saurin zama ma'aunin zaɓin mai siyarwa.

Allon Maki na Matakin Kwamiti

Allon maki mai amfani na kwamiti ga agentic AI ya kamata ya bi diddigin ma'auni shida — mafi ƙarancin saiti da ke banbanta shirin da aka shugabanta da wanda ba a shugabanta ba:

1. Rabon Matakin Cin Gashin Kai: Adadin workflow na gudanarwa bisa matakin (Level 0–4), da aka sabunta kowane kwata. Kowane workflow na Level-5 binciken da za a iya bayar da rahoto ne.
2. Cikar Control-Plane: Kashi-kashin workflow na gudanarwa masu dukkan sassa biyar na control-plane masu aiki (tantancewa, guardrails, policy-as-code, WORM logging, kill switch).
3. Cikar Audit-Trace: Kashi-kashin kiraye-kirayen workflow na Level-3+ masu cikakkiyar tushen da ake iya sake ginawa daga log da ba a iya canzawa. Manufa: kashi 100%.
4. Adadin Sokewa bisa Workflow: Kashi-kashin ayyukan da agent ya aiwatar da aka mayar baya cikin sa'o'i 24, da aka bi diddigi ga kowane workflow. Iyakar ƙararrawa: kashi 2%. Iyakar gargaɗi: kashi 5%.
5. Kuɗin Shawara na Yanzu: Kuɗin naúra na matakin workflow gami da kuɗin sokewa da gyara, idan aka kwatanta da tushen hannu. An bi diddigi a kan shari'ar tattalin arzikin shiri.
6. Yanayin Hujjar Tsari na Yanzu: Ranar sabunta hujjar tsari ta kowane workflow mafi kwanan nan a fadin tsarin da ya dace (SR 11-7, SS1/23, EU AI Act, MAS MGF). Kowane workflow da ya wuce kwana 90 daga zagayowar hujja binciken haɗari ne.

Waɗannan lambobi shida suna juya agentic AI daga slide deck zuwa tsarin aiki. Su ma su ne lambobin da mai binciken SR 11-7, mai dubawar PRA na wuri, ko hukumar kulawa ta EU za su fara nema.

Gibin da Wannan Ma'auni Ke Magancewa

Gibin tsari uku suna banbanta wannan ma'auni daga tsarin da ke akwai:

Gibi na 1: Ma'aunin da ke akwai suna auna girman AI, ba shugabanci na musamman ga agentic-AI ba. Evident AI Index yana auna Hazaka, Ƙirƙira, Jagoranci, da Gaskiya a fadin bankuna 50 ta amfani da bayanan da ake samu a fili. Ba ya — kuma ba a tsara shi ba — don tantance ko workflow na agentic na gudanarwa na banki na da kill switches masu aiki, WORM audit logs na kowane agent, ko OPA policy gates. Banki na iya zama na farko a Evident Index yayin da yake gaza binciken EU AI Act Article 12.

Gibi na 2: Tsarin tsari da ke akwai suna magana kan abin da ake buƙata, ba yadda za a auna shiri ba. SR 11-7, SS1/23, EU AI Act, FSB Sound Practices, da Singapore MGF kowanne yana bayyana wajibcin shugabanci. Babu wanda ke ba da tsarin auna mai fadin sassa da zai bar cibiya ta auna matsayinta a kan takwarorinta ko ta auna haɓaka cikin lokaci. Wannan ma'auni yana ba da wannan tsarin auna, ta amfani da tsarin tsari da ke akwai a matsayin tushen hujja.

Gibi na 3: Tattalin arziki na matakin shiri yana ɓoye gazawar matakin workflow. Mizanin masana'antar na ba da rahoton darajar AI a matakin shiri — "AI ya ceci awoyi X na aikin bin doka" — yana sa ba zai yiwu a binciko sokewa, fayil na SAR na ƙarya, ko aikin agent da ba a bayyana ba zuwa workflow da ya samar da shi ba. Sashen tattalin arziki na naúra na wannan ma'auni yana buƙatar alhakin matakin workflow. Wannan shi ne tsarin auna da ke sa tattaunawar CFO ta zama mai kariya da tattaunawar bincike mai tsira.

Ƙarshe

Agentic AI a bankuna a 2026 matsalar injiniya ce sanye da tufafin tattaunawar dabaru. Samfuri ana iya musanyawa. Control plane — OAuth scoping, semantic routing mai tabbas, OPA policy gates, WORM audit logs da ba a iya canzawa, da kill switch da aka gwada — ba a iya musanyawa ba. Tsarin shugabanci — tabbatarwar layi-uku-na-tsaro, suites na tantancewa na musamman na banki mai ci gaba, rahoton tattalin arziki na naúra na matakin kwamiti — ba a iya musanyawa ba. Kunshin hujjar tsari — SR 11-7 model cards na kowane workflow, EU AI Act Article 12 logs na kowane agent, taswirar FSB Sound Practice — ba a iya musanyawa ba.

Cibiyoyin da za su zama masu aminci ga masu kula da tsari a 2027 su ne waɗanda ke samun maki sama da 75 a fadin dukkan sassa shida na ma'auni a yau: suna rarraba kowane agent na gudanarwa a kan tsani na cin gashin kai, suna gina cikakken control plane mai sassa biyar, suna samar da hujjar tsari mai ci gaba, suna bin diddigin tattalin arziki na naúra na matakin workflow, suna zuba jari cikin shirin ƙungiya, kuma suna sa hannu da himma cikin tuntubar FSB, IOSCO, da masu kula da tsari na ƙasa da ke fasalta mizanan ɗaure na 2028.

OSWorld a kashi 66.3% shi ne iyakar dogaro. Tool-calls guda uku da aka haɗa a wannan matakin suna samar da nasarar kashi 29% daga farko zuwa ƙarshe. Ku shirya daidai. Cibiyoyin da ke auna agents yadda suke auna kowane haɗarin gudanarwa — ta hujja, ba burin ba — za su gano cewa shugabanci ba shi ne ƙuntataccen agentic AI ba. Shi kaɗai ne abin da ke sa agentic AI ya zama mai gasa.

Tambayoyin da Ake Yawan Yi

Menene bambanci tsakanin wannan ma'auni da Evident AI Index? Evident AI Index yana auna girman AI a fadin bankuna 50 na duniya ta amfani da bayanan da ake samu a fili a fadin Hazaka, Ƙirƙira, Jagoranci, da Gaskiya. Wannan ma'auni yana auna takamaiman tsarin injiniya da shugabanci — control plane, audit log, rarrabuwar matakin cin gashin kai, kunshin hujjar tsari — da ke sa agentic AI ya zama mai aminci don turawa a kan APIs na banki masu aiki. Ma'aunan biyu suna cika juna: Evident yana auna matsayin dabaru; wannan ma'auni yana auna shirin gudanarwa.

Wa ya kamata ya yi amfani da wannan ma'auni? Babban Jami'in Aiki, Babban Jami'in Haɗari, Babban Jami'in AI, shugabannin sarrafa haɗarin samfuri, da kwamitocin haɗari na kwamiti a bankunan duniya, bankunan yanki, ƙungiyoyin banki na kamfani, da cibiyoyin kuɗi da ke tura agentic AI. Haka kuma ya dace ga fintechs, PSPs, da masu siyar da kayan aiki da ke siyarwa cikin tsarin sayan banki inda hujjar tsari ma'auni ne na zaɓi.

Menene mafi ƙarancin matsayin shugabanci mai yiwuwa ga 2026? Cikakken control plane mai sassa biyar mai aiki cikin gudanarwa; dukkan workflow na gudanarwa an rarraba Level 0–4; an haramta workflow na Level-5 ta kwangila; WORM audit logs sun cika ga workflow na Level-3+; rubuta kowane agent na EU AI Act Article 12 a wuri kafin 2 Agusta 2026; FSB Sound Practices 1–4 an haɗa su da tsarin alhaki na kwamiti; suite na tantancewa na musamman na banki yana aiki mai ci gaba.

Menene sanarwar JP Morgan ke nufi ga cibiyata? Yana nufin ma'aunin gasa na turawar agent mai cin gashin kai yana da lokaci mai suna a 2026 daga banki mai muhimmanci ga tsarin duniya. Ba ya nufin kowace cibiya ya kamata ta dace da wannan lokaci. Yana nufin kowace cibiya ya kamata ta san makin ma'auninta na yanzu, ta san gibin da ke tsakanin wannan maki da matsayin turawa da JP Morgan ke bayyanawa, kuma ta sami ra'ayin da kwamiti ya amince da shi na zuba jarin shugabanci da ake buƙata don rufe wannan gibi cikin aminci.

Yaya ya kamata a ba da rahoton haɗarin agentic AI ga kwamiti? Ma'auni shida ga kowane workflow: matakin cin gashin kai, cikar control-plane, cikar audit-trace, adadin sokewa, kuɗin shawara na yanzu, da yanayin hujjar tsari na yanzu. Tare da jerin manyan biyar na ragowar haɗari. Ku bar slideware na model-card da taƙaitattun ribar matakin shiri.

Shin tuntubar FSB tana haifar da wajibci masu ɗaure yanzu? A'a. FSB ta bayyana a sarari cewa Sound Practices 12 ba mizanai masu ɗaure ba ne. Sai dai, tuntubar tana rufewa a 22 Yuli 2026 kuma rahoton ƙarshe yana zuwa ga ministocin kuɗi na G20 a Oktoba 2026. Masu kula da tsari na ƙasa — Fed, PRA, BaFin, DNB, ACPR, MAS — suna da 'yancin haɗa Sound Practices cikin tsammanin kulawa masu ɗaure a kan lokutansu. Cibiyoyin da ke mayar da martani ga tuntubar yanzu su ne waɗanda ke fasalta yadda mai ɗaure zai yi kama.

Manazarta

1. Cambridge Centre for Alternative
2. Report finds uneven AI adoption in financial services - News & insight
3. The 2026 AI Index Report
4. FSB Issues Consultation on Sound Practices for Responsible AI ...
5. Sound Practices for Responsible Adoption of Artificial Intelligence (AI)
6. Kakunin Compliance Angle
7. The EU AI Act Compliance Deadline Is août 2026: What Financial Services Firms Need to Do Now
8. JPMorgan Chase plans to deploy more powerful AI agents this year
9. JPMorgan Chase to deploy long-running autonomous AI ...
10. Evident AI Index
11. McKinsey's latest report on agentic AI in banking found that
12. Singapore Launches New Model AI Governance Framework for ...
13. Singapore's Agentic AI Framework: The Most Forward-Looking AI ...
14. Financial Stability Board points banks towards AI monitoring AI as ...
15. [PDF] The 2026 Global AI in Financial Services Report
16. CCAF AI-Monica Jasuja - LinkedIn
17. Key findings from the 2026 Global AI in Financial Services Report by the University of Cambridge
18. AI Use Case Trends in Banking
19. AI governance gap widens as enterprises race to deploy agentic AI ...
20. Agentic AI will shake up banking, shrinking global profit pools
21. Precision, Not Hype, Will Shape Banks' Use Of AI In 2026 - Forbes
22. 2025 Evident AI Banking Index: Who's Leading in AI? - Teradata
23. Agentic AI Banking Strategy: A C-Suite Planning Guide
24. Agentic AI: Banking's Next Frontier Beyond the Chatbot - CCG Catalyst
25. Singapore Introduces New Model AI Governance Framework for ...
26. IOSCO sets out supervisory framework for AI use in capital markets
27. Publication of IOSCO AI Supervisory Toolkit and Industry Practices ...
28. Best Practices for AI Governance and Risk Management Published ...
29. Banking’s agentic AI opportunity
30. IOSCO Publishes AI Supervisory Toolkit for Capital Markets - LinkedIn
31. Agentic AI Governance in Banking: Closing the Gap in 2026
32. EU AI Act Compliance for Financial Services: Complete 2026 Guide
33. MAS and AI in Singapore Financial Services - AIRiskAware
34. MAS Releases AI Governance Framework Version 2 for Singapore Financial Services — AIMenta
35. De Autonome Treasury-Index in 2026: agentische treasury en programmeerbare liquiditeit — Sebastien Rousseau

An sake bita ta ƙarshe 2026-06-30.

Bita ta ƙarshe 2026-06-29.

# Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau

> Originally published at [https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/](https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/)

Ma’auni mai bangarori shida don tantance shirin bankuna ga agentic AI: autonomy, governance, shaidar doka, tattalin arziki, shiri, da daidaiton duniya.

Read the full article on sebastienrousseau.com: https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/

Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau

Ma’auni mai bangarori shida don tantance shirin bankuna ga agentic AI: autonomy, governance, shaidar doka, tattalin arziki, shiri, da daidaiton duniya.

https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/

Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau

Ma’auni mai bangarori shida don tantance shirin bankuna ga agentic AI: autonomy, governance, shaidar doka, tattalin arziki, shiri, da daidaiton duniya.

Ga abubuwan da ya kamata a lura da su na dabarun:

- Me Yasa Wannan Ma'auni Ke Wanzuwa. Evident AI Index yana matsayin bankuna 50 na duniya a fadin Hazaka, Ƙirƙira, Jagoranci, da Gaskiya ta amfani da miliyoyin bayanan da ake samu a fili.
- Yanayin Girman Agentic AI na 2026. Rahoton Cambridge CCAF na 2026 — babban binciken duniya na AI a fannin kuɗi, da ya shafi ƙungiyoyi 628 a fadin hukunce-hukunce 151 cikin haɗin gwiwa da BIS, IMF, WEF, da Bankin Duniya — yana ba da tushen ƙididdiga na…
- Tsarin Ma'auni Mai Sassa Shida. Wannan ma'auni yana auna shirin agentic AI a fadin sassa shida.
- Makin Ma'aunin Haɗaɗɗe. Makin sassa shida suna haɗuwa zuwa ma'auni haɗaɗɗe ta amfani da auna nauyin-mahimmancin-tsari mai zuwa:.

Menene hanyar ƙungiyar ku wajen magance ƙalubalen da aka kawo a wannan rubuce-rubucen?

→ https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/

#AgenticAi #AgenticAiBanking #AiGovernance #AutonomyTiers #ModelRiskManagement

Sebastien Rousseau | CC-BY-4.0

@online{rousseau2026ma,
  author  = {Rousseau, Sebastien},
  title   = {{Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau}},
  year    = {2026},
  url     = {https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/},
  urldate = {2026}
}

TY  - GEN
AU  - Rousseau, Sebastien
TI  - Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau
PY  - 2026
UR  - https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/
ER  -

Rousseau S. Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau. sebastienrousseau.com. 2026 Jun 30. Available from: https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/

Rousseau, Sebastien. "Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau." sebastienrousseau.com. June 30, 2026. https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/.

Rousseau, S. (2026, June 30). Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau. sebastienrousseau.com. https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/

Ma’aunin Agentic AI ga Bankuna a 2026: Auna Autonomy — Sebastien Rousseau

Ma’auni mai bangarori shida don tantance shirin bankuna ga agentic AI: autonomy, governance, shaidar doka, tattalin arziki, shiri, da daidaiton duniya.

Originally published at https://sebastienrousseau.com/ha/2026-06-30-agentic-ai-index-banks-measuring-autonomy-2026/ by Sebastien Rousseau.
Licensed under CC-BY-4.0.

SEBASTIEN ROUSSEAU · FOUNDER · ENGINEER