Issue 6
Viewpoint
supervision · global
BIS, Bank of England and central banks warn of AI-driven financial stability risks
The head of the BIS warned on 10 September of new financial stability risks from the AI boom. In a letter of 28 August, Bank of England Governor Andrew Bailey also warned G20 finance ministers and central bank governors about the risks of frontier models.
Who is affected and how
- chief financial officer
- Risk models should carry a separate item for dependence on AI providers and the danger of fast, correlated decisions.
tool · United States
OpenAI launches ChatGPT for Financial Services, aimed at Wall Street junior bankers
OpenAI launched ChatGPT for Financial Services on 10 September, Reuters reported. Hungarian reports say the product could also affect the work of junior staff on Wall Street.
banking professionalThey need to assess which junior analyst and preparatory tasks the new product can replace, and where human review is needed.chief financial officerIt is worth starting a pilot and comparing the price with the cost of in-house AI development.fintech companyA major player's own financial product has to be factored into competition and partnership strategy.tool · globalAnt International, Visa and Mastercard agree on common identity standard for AI agents
Ant International said on 9 September that it is working with Visa and Mastercard on developing AI payments. The companies agreed on a common standard for identifying AI agents, but implementation is…
fintech companyAgentic payment products will have to be aligned with this identification standard, or they risk being left out of the networks.banking professionalBanks need to prepare to identify, authorise and handle disputes for transactions initiated by agents.retail customerThe prospect of an AI agent paying on a customer's behalf is approaching, so the limits of the mandate need to be set in advance.fraud · globalOrganised fraud rings use deepfakes and reused identities to target banks
Reports say organised groups are using deepfakes and reused identities, while banks in Asia-Pacific also face a new threat from synthetic identities. AI is also changing the economics of fraud farms…
banking professionalAccount-opening and customer-identification screening must be strengthened against patterns of synthetic and reused identities.fintech companyFraud-prevention development must be scaled to cheaper, mass-volume attacks.regulation · AsiaIndia prepares AI-agent registry and Unified Agent Protocol for UPI
At Global Fintech Fest, NPCI presented the Unified Agent Protocol (UAP), which would govern the registration, identification and authorisation of AI agents in UPI payments. According to a Reuters…
fintech companyUPI agents will only be able to operate in a registered, verified way, which developers need to factor in now.banking professionalBanks will need to verify agent authentication and handle authorised transactions.Hungary
surveySurvey: a quarter of Hungarians discuss their own finances with AIretail customerAI advice needs to be checked separately, as it does not replace personalised, responsible financial advice.banking professionalCustomers often arrive after an AI conversation, so advice has to respond to the opinion they have already formed.insurerInsurers should expect customers to bring a preliminary view formed through AI when requesting quotes and asking for product explanations.deploymentAI is live in about 170 places at Hungarian banks; payback and the human factor are the questionsbanking professionalAI pilots need to become dedicated projects with measurable payback, starting from staff use.chief financial officerInvestments should be tied to payback indicators, because AI speeds up long planning cycles.regulationCentral bank makes AI a strategic priority; AI governance is the sector's taskbanking professionalFor AI in credit assessment, systems should be designed around the MNB's expectations and explainability.fraudCzech and Slovak reports: scammers target people, with AI supportbanking professionalCustomer education and transaction warnings should be tuned to psychological manipulation and mobile-app attacks.retail customerDo not install an app received after a phone call or message, and do not approve unfamiliar requests.fraudHungarian trade press: AI has opened a new era in fraud and makes it cheaperbanking professionalInternal payment-authorisation processes should require callback verification across multiple channels.chief financial officerDual approval should be introduced against transfer fraud based on executive impersonation.retail customerVoice and video alone no longer prove a person's identity.caseClaude, ChatGPT and Grok went down for an afternoon; none is listed among banks' critical providersbanking professionalA fallback procedure is needed for when the AI assistant goes down.deploymentGránit Bank and BME sign cooperation agreement, with AI ideas competitionbanking professionalHungarian banks are building AI talent and ideas through cooperation with universities.caseHungarian press: AI agents may haggle over prices and pay on users' behalfbanking professionalFor domestic payment products, the handling of agent-initiated transactions should be planned in advance.retail customerA predefined limit should be set on the payment authority given to an agent.surveyCorvinus University and UNIQA research how companies adopt AIinsurerThe research findings could offer lessons on work organisation for AI adoption in insurance.caseOminimo, a Hungary-launched AI-based car insurer, is among nominees for fintech decadeinsurerA Hungarian insurtech example showing that cross-border scaling is possible with AI-based pricing.fintech companyThe Hungarian ecosystem's success story is a reference for investor and expansion strategy.
More key stories of the week
supervision · United Kingdom
FCA frontier AI review: financial firms inundated with AI-generated security findings
The FCA's frontier AI review and warning say financial firms are being overwhelmed by a flood of AI-generated security findings. The material published in the first week of September also bears on…
banking professionalIT security teams must plan capacity to handle vulnerabilities uncovered with AI tools.tool · globalAnthropic unveils agentic commerce blueprint backed by Visa and Mastercard
Anthropic launched the Agentic Commerce Blueprint on 7 September, backed by both Visa and Mastercard.
fintech companyPayment and commerce products should be aligned with the agentic pattern supported by the card networks.banking professionalBanks need to assess how purchases initiated through an AI assistant appear at the issuing bank.tool · globalCoinbase unveils AI financial accounts and limited wallets as x402 spreads
Coinbase unveiled AI financial accounts and limited wallets on September 9. The x402 HTTP payment protocol had reportedly reached 165 million cumulative transactions and roughly $50 million in volume…
fintech companyAgent wallets with limited permissions offer a template for designing spending limits.case · globalInsurers are still working out how to price and protect against AI risk
Between September 8 and 13, several analyses addressed insuring AI liability: pricing is difficult, new risks may bring exclusions, and standalone AI cover may emerge. Beazley says digital health…
insurerContracts with AI exposure need to spell out what is covered and excluded, and the case for standalone AI cover should be assessed.chief financial officerThe corporate insurance programme should be reviewed to see whether losses from AI decisions are covered.case · United StatesJPMorgan curbs lending to Situational Awareness after AI losses, source says
JPMorgan has reduced its lending to Situational Awareness after the firm suffered losses on its AI positions, according to a source. The news was reported on September 11.
banking professionalExposure limits on financing and collateral management for AI-themed investment funds should be reviewed.chief financial officerLender and counterparty exposures tied to AI should be stress-tested.survey · globalThe AI return question: bank FOMO, CFO budgeting and a modest impact on results
An Accenture executive says banks' AI spending is driven by fear of falling behind. Analyses say CFOs are applying the SaaS budgeting playbook to AI, and McKinsey reports that 37 percent of companies…
chief financial officerAI budgets should be built on usage-based costs and measurable outcomes.banking professionalInvestment requests should be tied to a clear business metric.survey · globalMastercard report: one in ten online shoppers could buy and pay through AI agents by 2030
A Mastercard report published on 8 September says that by 2030 more than one in ten online shoppers could regularly use AI agents to shop. In the survey of 26,000 respondents, trust still stops…
fintech companyFor product planning, it is a capacity estimate that nearly one in ten consumers may pay through an agent.banking professionalBanks need to prepare for agent-initiated card transactions and for handling customer authorisation.retail customerUsing a shopping assistant and handing over payment authority are separate decision steps.deployment · European UnionRevolut and Visa complete France's first AI agent-initiated payment
Revolut and Visa announced on 7 September that they had successfully completed the first AI agent-initiated payment in France. According to the Hungarian press, the agent made the purchase with a…
banking professionalThe first experience of how agentic payments work on cards, and of customer approval, is worth analysing.fintech companyAgentic products from neobanks and payment providers mean either a competitive advantage or competitive pressure.tool · globalMastercard launches Agent Connect for merchants; Visa Agent Pay is the other side
Mastercard announced Agent Connect on 9 September, which helps merchants prepare for AI shopping. Together with Visa Agent Pay, the two networks are competing over the rules of agentic shopping.
fintech companyOn the merchant and acquirer side, firms must decide which network's agentic solution to join.banking professionalIssuers need to prepare to handle the two networks' agent tokens and identification.layoffs · AsiaAxis Bank turns to AI to slow hiring and grow with a flat headcount
Axis Bank is relying on AI and will not lay off staff but will hire new employees more slowly, according to reports of September 9–10. The Indian bank aims to grow with a flat headcount.
banking professionalEntry-level roles are shrinking, so training should focus on supervisory and review work alongside AI.investment · United StatesBanks seek lead roles in Anthropic IPO and top credit ratings for Anthropic and OpenAI
Morgan Stanley and Goldman are set for leading roles in Anthropic's IPO, according to The Banker. According to the Financial Times, bankers advising Anthropic and OpenAI want the companies to secure…
banking professionalIPOs and credit ratings of AI labs open up a new client base for transactions and risk business.chief financial officerInvestment portfolios should track AI exposure and the ratings process.regulationAPRA and ASIC issue coordinated direction on frontier AI risk
Australia's prudential regulator APRA and securities regulator ASIC acted in a coordinated way on the risks of frontier AI. Legal analyses set out what financial service providers need to do.
banking professionalFor services built on frontier models, risk management documentation needs to be checked.In brief
regulationTelangana tax department holds officers personally accountable for unauthorised AI sharing of taxpayer dataaccountantClient data may be entered into a public AI tool only in line with internal policy and data-protection approval.regulationSingapore to pilot common prefix for government calls and cross-bank AI against scamsbanking professionalCross-bank data sharing allows faster identification of suspected scam accounts, and the model can be adopted elsewhere.retail customerA uniform calling prefix helps tell genuine official calls from impersonation.oversightRBI warns of speed, concentration and opacity risks in AI use in financebanking professionalFast, automated decisions need built-in human review points and a shutdown mechanism.caseMoody's warns banks over their dependency on AI technologychief financial officerDependence on third-party AI providers must also be addressed in credit rating and risk communication.regulationFTC now targets payment processors that enable AI-driven fraudfintech companyPayment processors need to tighten merchant screening and the flagging of suspicious traffic.banking professionalIn acquiring-side risk management, responsibility grows for screening out fraudulent merchants.toolVisa launches AI fraud shield meant to stop scams before money leaves accountsbanking professionalBanks should assess how pre-payment risk scoring could fit into their own fraud-screening systems.fintech companyFor acquiring and issuing services, the Visa tool may be a competitor or an integration opportunity.retail customerPre-payment checks may mean delays or requests for confirmation on suspicious transactions.regulationGerman legal analysis: does AI agents' payment processing require a licence under the ZAG?fintech companyAn agentic payment business model should be reviewed for licensing requirements before development begins.toolEsker, Paystand, cfo.ai and SoVa launch agentic AI tools for finance teamschief financial officerAutomation of invoicing, receivables and planning tasks should be tested in a pilot, with review points built in.accountantRoutine tasks move to agents, and the emphasis shifts to review and client communication.caseInsurers: AI in claims and underwriting, and a decline in insurance jobsinsurerIn claims and underwriting, the time saved by automation should be weighed against the cost of human review.chief financial officerThe reason for headcount reductions should not be attributed solely to AI.fraudDeepfake investment scams: Portuguese warning, Northern Irish and Indian victimsretail customerVideo investment offers from celebrities should not be believed, and remote access to a computer should not be granted.banking professionalFor large investment transfers, a delaying verification question is worth applying.toolFraud-prevention vendors: ITSS and Feedzai partner, Darwinium wins awardbanking professionalNew, award-winning players have appeared in the APP fraud solutions market and are worth comparing.fintech companyThe partnership model is one route for spreading AI fraud prevention.deploymentIndian finance apps add AI agents: Super.money and L&T Finance with Pine Labsfintech companyIn consumer apps, agent-based shopping can be a competitive edge against larger rivals.banking professionalAgent-based payments in NBFC and app environments also affect transaction risk management.deploymentKIWI Finance launches closed-environment agentic AI ecosystem for lending, built on FlowX.AIbanking professionalThe closed-environment agent system is a model for automating the lending process and managing its risk.investmentAutonomous AI-agent hedge fund and JPMorgan agents' backtest resultschief financial officerBacktest results do not prove live performance, so agent-based strategies need separate verification.caseVisa CEO: AI shopping has arrived, but agentic payments haven'tchief financial officerRollout plans are more realistic if they assume a slower pace of agentic payment adoption.fintech companyThe task is to resolve the trust barriers to handing over payment, not just the technology.deploymentCanada's big banks use AI to recover loans and cut call-centre volumes as workers face uncertaintybanking professionalCollections and customer-service roles are changing, and skills development should be shaped accordingly.chief financial officerMeasuring the claimed savings and accounting for the cost-side impact is the job of decision-makers.toolTrue Balance scores every loan application with AI, using alternative databanking professionalIncluding alternative data can sharpen the assessment of customers without credit bureau profiles.investmentKPMG takes minority stake in deepfake detection firm Reality Defenderchief financial officerA separate budget line for impersonation-fraud controls is worth planning.toolCaseware brings agentic AI into audit workflows and Microsoft ExcelaccountantIn audit work, the agent's output should be used alongside professional judgment and documentation.toolIndonesia develops AI system to detect unpaid corporate taxes; Kyrgyzstan plans AI in tax administrationaccountantAs tax authorities apply AI-based data analysis, companies' records and returns need to be consistent.chief financial officerA tax-risk review is advisable, as detection by the authorities is becoming automated.toolExperian launches AI-enabled real-time decisioning platform in its consumer marketplacefintech companyLoan brokers and issuers should look into the possibility of integration.banking professionalCash-flow data can be a supplementary input for underwriting models.fraudFounder and former CEO of an AI company pleads guilty to securities fraudchief financial officerAI providers and AI-investing targets should be subject to stricter scrutiny.fraudBank Negara again warns of deepfakes impersonating its governorretail customerInvestment appeals that claim to come from the central bank or its governor should not be accepted unless verified through official channels.toolPaytm to enter AI business, plans to sell agents to banks and insurersfintech companyPayment providers can enter the banking and insurance software market with AI agent products.banking professionalOn the vendor side, a new procurement alternative is emerging for voice-based and process-automation agents.surveyKPMG and Forrester: agentic commerce attracts fintech investment, with B2B the proving groundfintech companyIn B2B payments, it is worth looking for agentic use cases.chief financial officerIt is worth preparing to make corporate payment processes agent-friendly.investmentChime to acquire Stride Bank, ending its sponsor-bank modelfintech companyNeobanks are increasingly moving from the sponsor-bank model toward holding their own banking licence.banking professionalThe reshuffling of sponsor-bank relationships also affects partner banks' business.regulationAI fraud pushes US toward cryptographic digital identitybanking professionalProcesses built on document-based identification should be reviewed.deploymentM&T Bank rolls out AI copilots to more than 15,000 employeesbanking professionalAlongside copilot use, a regulated scope of use and training should be planned.deploymentIncore Bank trials Gemini-powered KYC with Kyndrylbanking professionalThe time saved through KYC automation is worth measuring in the customer onboarding process.investmentSilver Lake to merge Cegid and Silae in 10 billion euro dealaccountantThe accounting and payroll software market is consolidating, so it is worth assessing exposure to suppliers.chief financial officerFinance and HR software may end up on a single platform, which affects vendor-switching decisions.caseplaceholdercaseplaceholder