AI Infrastructure Lending Reform and Digital Banking Expansion

Taiwanese banks collaborate to build AI

At a cost of around $2 million, a project by Taiwan’s bank to build a local AI service seems like a bargain – and could provide a template for other countries looking to build AI for financial services. Banks have long used specialist AI and machine learning internally for credit scoring and security, rather than the generative AI chatbots such as ChatGPT that have surged in popularity in the last three years. Sixteen banks are set to collaborate to deliver a service specifically for Taiwan. “The group aims to build an AI infrastructure ‘best suited to Taiwan’s financial industry,’ according to a statement released late Wednesday,” reports Bloomberg. “The project will be led by CTBC Financial Holding Co and overseen by the Financial Supervisory Commission. The envisioned LLM is based on an open-source model but not Chinese platforms. The model will be trained on specialized datasets including industry-specific data and regulatory frameworks provided by the FSC. It will use the Digital Ministry’s Sovereign AI database to ensure the model reflects local linguistic and cultural contexts.” Training for the model will begin soon with the goal of having it ready by next year. “CTBC Bank Chairman James Chen said that a divide is emerging between ‘AI-ready’ companies and laggards, with the former seeing higher revenue growth and market valuations.”

US mortgage giants update their lending models

Freddie May and Fannie Mac, the giant banks that guarantee US mortgages, will now accept credit scores that take into account rent payments and other bills. The Trump administration has promised to make mortgages more accessible. “Federal Housing Finance Agency Director William Pulte said the pair, which guarantee most U.S. mortgages, will now accept mortgages assessed using the VantageScore 4.0 model, which uses advanced data and analytics to generate scores,” reports Reuters. “He said they are also working to implement a similar modified score system from FICO. The officials said the expanded use of other credit score models should boost competition and potentially lower mortgage costs, allowing opportunities for potential homebuyers who otherwise would have struggled to gain an affordable mortgage. ‘We are modernizing credit scoring with more predictive models, helping millions of Americans who responsibly pay rent qualify for mortgages.  

That’s fair, it’s commonsense, and it’s finally delivering the benefits of competition to homebuyers nationwide,’ said Pulte in a statement.” VantageScore 4.0 was released in 2017, is offered by the three main credit reporting agencies Transunion, Equifax and Experian and is only available in the US.

Kenya reins in digital lenders

Kenyan lenders will be required to prove borrowers are capable of repaying loans before giving approval under new laws proposed by regulators. Digital lenders aka digital credit providers such as Tala and Branch have seized market share with instant approvals and credit rated on behavioural data. “The proposed rules would fundamentally change how credit works in one of Africa’s most active digital lending markets,” says TechCabal. “Kenya has more than 227 licenced digital credit providers, but default rates have drawn sustained regulatory concern. The framework is the regulators’ attempt yet to address a market that expanded faster than the rules governing it”. A draft of the new rules appears in a Financial Consumer Protection Framework issued last month by the Central Bank of Kenya, the Capital Markets Authority (CMA) and the Communications Authority of Kenya (CA), and would apply to banks, fintechs, and mobile money providers. “‘A Financial Services Provider (FSP) shall not provide a credit product… unless they have first undertaken a reasonable assessment to confirm the retail consumer’s ability to repay the credit without financial hardship,’ the draft noted.” Central bank data, says TechCabal, showed default rates on small loans of up to 80 per cent. “Larger digital loans perform better, but overall default rates for digital lenders have been reported as high as 40%, more than double those in the banking sector. Digital lenders typically approve loans using alternative data such as mobile money transactions, airtime usage and device metadata, with decisions made in seconds and little verification of income or expenses.”

TikTok’s baffling payment system

For many small business owners or “solopreneurs”, it’s a challenge to separate one’s private life from one’s business. It’s even harder for influencers and streamers, who are “always on”. TikTok, one of the main platforms for streamers, has created a debit card to help creators better separate business and personal finances, reports Finextra. Getting paid on TikTok is complex. A successful content creator doing a TikTok Live earns virtual income through a TikTok tokens ecosystem. Viewers purchase TikTok Coins (which are virtual coins) with real money, and use these TikTok Coins to purchase Virtual Gifts, which they can use to reward creators they enjoy. TikTok takes about 50 percent of these gifts as a fee. Creators then convert the remaining gift value to TikTok “diamonds”, another TikTok virtual currency. Once creators have earned the equivalent of $100, they can then convert the “diamonds” into local currency, minus a 50 per cent fee to TikTok.  “According to a survey carried out by the payments giant of creators on multiple social media platforms, 86% of creator-run businesses are self-funded and 49% experience late payments, highlighting the ongoing cash flow challenges they face,” writes Finextra. Visa says the card will help creators manage cash flow better by making access to funds simpler.

Plata’s value is growing fast in Mexico

Latin American digital bank Plata has raised Series C funding of $400 million in a round that drew in investors ranging from Qatari wealth funds and Brazil’s BCG Pactual to US university endowments, doubling its valuation in three months to $5 billion. The three-year old business acquired a digital banking licence in Mexico last month. Plata has selected strategies that have been successful elsewhere. For a start, its core team came from Russian bank Tinkoff, but it has added the credit-card strategy of Nubank, activating 3.5 million credit card customers, a quarter of whom are using their first credit card. The digital bank delivers the cards in person, and gets over 40 per cent of its business through word of mouth.

In three years of operation, the company has surpassed 3.5 million active credit card customers, over 750,000 of whom received their first-ever credit card. Its top-rated app, extremely responsive customer service, and innovative ambassador-driven onboarding drive unparalleled virality, with over 40% of customers coming through referral and organic channels. “The launch of deposit and debit capabilities now allows Banco Plata to bring full-service digital banking to Mexican consumers for the first time,” reports financialit.com. “Plata reached over $600 million in annualized revenue and an $800 million loan portfolio, powered by proprietary AI risk models and platforms built by over 800 STEM professionals. In July 2025, Plata also received authorization to incorporate a Compañía de Financiamiento in Colombia.” Chief executive Neri Tollardo previously told Lafferty News that his strategy was to hire data scientists over traditional bankers. Plata built its core banking system and its own CRM system. "We built a technology-led platform designed to broaden access to better financial services at scale,” said Tollardo. “The launch of full banking operations in Mexico is a pivotal milestone – it expands our product range and gives us access to retail deposit funding, meaningfully strengthening our funding model for the next phase of growth."

Mythos gets free advertising from central banks

A rolling wave of bank regulators is raising alarms about the new Mythos AI model from Anthropic. (One imagines criminal masterminds sending their lackeys to get their hands on a copy.) First it was the Americans, whose major bank leaders were summoned to the White House two weeks ago. This week, it’s Asian and European regulators. “The Hong Kong Monetary Authority (HKMA) said it has engaged with major banks and is highly vigilant to the evolving AI-driven cyber threats – such as Mythos – and is about to bring forward a new framework to address the challenge,” reports Reuters. “HKMA will also form a new dedicated public-private sector taskforce to examine, monitor and respond to AI-driven cyber risks.” Regulators from Australia, South Korea and Singapore also raised alarms this week about Mythos.

German central bank chief Joachim Nagel said banking authorities must act to prevent the misuse of Anthropic’s most advanced AI model to date, as it opens the door to new and sophisticated cyber risks. “‘This AI model seems to be a double-edged sword, since it could be used not only to improve digital security systems, but also to leverage their vulnerabilities for malicious purposes,’ Nagel said at an event in Rome.”