North Korea’s TraderTraitor group drained about US$6.4 billion from Binance in the early hours of 30 October, Singapore time, according to a Chainalysis reconstruction that puts the loss slightly above the exchange’s own US$6.3 billion estimate. No cryptography was broken. A compromised engineer’s laptop let the attacker obtain valid signed authorisations from a Binance hot wallet, and the exchange’s systems did exactly what they were told.

What followed was a race against the freeze button. The attacker gave up roughly US$40 million converting stablecoins into ETH, but freezes began 54 minutes after the first outflows. Tether and Circle account for about US$2.4 billion of the US$3 billion Binance says is now frozen. Roughly US$2.4 billion in unfreezable assets is still moving through thousands of scripted Tron transactions, a pattern Chainalysis says it has not seen before.

Binance resumed withdrawals in phases on 1 November, said it had covered 95% of stolen assets and recorded a US$3.6 billion net outflow in the first nine hours. For Southeast Asia, where stablecoins are becoming everyday infrastructure, the uncomfortable lesson is that the rescue depended on a handful of issuers based outside the region.

—

REGIONAL

Krungsri joins Amity’s Series D as the Thai AI group eyes a 2027 IPO: The undisclosed cheque from Krungsri Finnovate’s fund extends a US$100M-plus round led by EDBI. A preliminary data partnership with Amity’s EGG Digital is the bigger story, though neither side has said what customer data will be shared.

Taiwan and Singapore state funds back NSV’s US$53M debut biotech fund: NeoSci Ventures’ first close lands as first-time venture fund closings fell nearly 80% from 2021. Two of four named LPs are state vehicles, so the undisclosed final close will test private appetite.

Binance seeks a licensed route back into Malaysia after its 2021 ban: The exchange says it has spent 10 months talking to the Securities Commission and Bank Negara Malaysia. The regulator recently put Binance back on its Investor Alert List, and the talks overlap with its record hack.

Google funds US$160M water plant to serve Negeri Sembilan data centre: Gamuda will build a reservoir and treatment plant supplying 65 million litres a day of potable water, with spare capacity for public use. Google says its assessment found local sources face no high scarcity risk.

Vietnam’s climate-tech funding halves as mid-sized cheques vanish: No recorded deal closed between US$350,000 and US$1M in 2025. Women founders won 56% of grant funding but received only about 1.7% of total capital, partly because banks demand land titles they rarely hold.

Vynn Capital backs Econili, taking its total funding to US$1.97M: The Ipoh battery recycler will spend it on feedstock and throughput. The sum looks modest beside its 24,000-tonne plant, a planned Spanish facility and Nasdaq ambitions, and Econili has not said how it funds them.

10 Cents Capital raises US$20M to launch its own ventures in Brazil: The Malaysian investment holding firm is moving from backing companies to running its own, across AI business applications, multilingual marketing and cross-border B2B trade. Brazilian brand VIVAMOMENTO will handle local operations.

Singapore’s HugoHub goes live with Pakistan’s digital bank HugoBank: The pilot runs under a restricted State Bank of Pakistan licence, with NADRA biometrics and Raast payments built in. HugoBank seeks commercial approval in early 2027; only about a quarter of Pakistani adults held an account in 2024.

Multipolitan raises US$2M seed at a US$25M pre-money valuation: EmergentX led the round for the Singapore-registered firm, which bundles visa, immigration and travel admin. Its Private Client arm claims US$4M-plus in net revenue in under two years; the consumer app launches in December.

INTERVIEWS & FEATURES

Roblox’s new SEA chief Joel Neoh must win over regulators before users: The Fave founder starts as Indonesia’s PP Tunas rules and Malaysian scrutiny put child safety first. Brand tie-ups such as Marrybrown’s loyalty experiment will test how far he can push commerce on a young platform.

AmanTrue CEO says AI agents need passports, not just exams: Evin McMullen argues Singapore’s accredited testing covers competence before launch but not accountability afterwards. She stayed vague on paying customers and revocation speed, and admits a commercial interest in mandatory agent identity.

INTERNATIONAL

OpenAI cuts its round target to US$20B after a tepid investor response: UAE investors will lead, and the WSJ reports US$11B is secured so far. The retreat from a US$30B ambition comes as Wall Street sours on heavy AI infrastructure spending ahead of planned listings.

Theranos whistleblower warns the AI boom makes due diligence harder: Tyler Shultz says investors struggle to tell genuine innovation from a ChatGPT wrapper, which invites deception. Elizabeth Holmes’s partner already runs diagnostics startup Haemanthus, which has reportedly raised about US$20M.

CYBERSECURITY

Hackers defrost European supermarket freezers in a second wave: Stores in Belgium, Ireland, Luxembourg and Sweden lost stock after attackers changed settings on Danfoss controllers that no longer get security updates. US officials are probing similar freezer hacks on military bases.

SEMICONDUCTOR

Malaysia’s chip exports surge, but its market share hasn’t budged: CGS International says memory exports rose 243%, slightly below global memory growth of 249%. E&E wages and jobs are flat, and quarterly investment approvals sit far below the 2021–2023 peaks.

Apple’s inventory jumps 154% as it stockpiles costly memory chips: Inventory hit US$14.54B, and Apple guided December-quarter margins down to 45–46%. CEO John Ternus called memory pricing unprecedented and expects costs to keep rising, while purchase obligations climbed US$33B in a quarter.

AI chip demand keeps Singapore manufacturers upbeat into 2027: A net 24% of manufacturers expect better conditions to March 2027, led by precision engineering and electronics firms serving AI data centres. Chemicals is the gloomiest cluster, citing crude feedstock uncertainty.

AI

Musk throttles Grok Bot as serving costs outrun revenue: Users found weekly limits maxed out without warning over Halloween weekend. The cut sits oddly beside SpaceX’s claim that AI makes up 52% of income and that its AI division turned a US$431M profit.

YC’s Garry Tan spends up to US$200K a year on AI agent tokens: Tan says personal agents will go mainstream only if costs fall, calling his spending a privilege. YC also runs an internal agent holding the context of every founder and application it has seen.

Privacy-first AI assistant Ollie shuts down two months after its seed: The Khosla-backed startup raised US$7.5M and won SOC 2 compliance, but founder Bill Lennon said rivals made it too hard to stand out and raise more capital. The app closes on 6 November.

Apple’s Ternus says the company won’t ship AI for its own sake: The new chief executive told Wired he likes Apple’s AI position despite Siri delays. He dismissed the idea that agents could make apps unnecessary, calling product creation deeply human.

THOUGHT LEADERSHIP

Winning at home no longer satisfies SEA’s venture investors: Yvan Goudard argues founders should design for a second market early without sprawling. The best next country shares customers, regulators or language, and lowers the cost of entering a third.

The Singapore–Romania corridor needs introductions, not more MOUs: Aleks Farseev says Romanian founders pitching to Singapore asked for warm contacts rather than grants. Corridors should be judged by meetings and first customers, with chambers owning named follow-ups.

AI speeds up decisions, but owning them still caps scale: Drawing on four businesses, Fathhi Mohamed argues coordination load grows with every market. Firms should separate recommendation from authority in writing and build talent density before buying tools.

Bitcoin now trades on elections, oil and missiles, not crypto news: Anndy Lian points to Bitcoin’s 83.8% correlation with the S&P 500 as US midterms and strikes on Iran’s Qeshm Island hit risk appetite. He calls US$76,000 the line in the sand.

Bitcoin’s rebound looks like a short squeeze, not a reversal: Anndy Lian says short liquidations drove Bitcoin’s 1.16% gain to US$78,293 after Binance reopened withdrawals. Ethereum followed as a beta trade, and Bitcoin must reclaim US$79,107 to convince.

AI fears deserve independent evidence, not founders’ word: Ajay Tiwari says warnings from Sam Altman, Dario Amodei and Geoffrey Hinton merit scrutiny, as fear also brings influence. He backs independent pre-deployment safety tests and rules matched to actual risk.

Confident AI answers still need checking, says a publisher: Melissa Fann of The Wellness Insider recounts ChatGPT misreading Singapore TOTO results and forgetting instructions in long projects. Her rule: the cost of a mistake decides how much you verify.

Inside MAIE, the AI engine behind a Dubai policy workshop: Alex Lin explains how prompts demand what sources leave out, pairing every forecast with a ‘Not Quite Right’ tripwire. Humans still choose the readings, the speakers and the projects.

Children want to read stories their peers write, Panel One finds: Anumeha Bisaria’s competition drew 537 entries from 325 schools across 10 countries. AI tools turned stories into graphic novels, but mentors kept students revising.

 

The post Ecosystem Roundup: North Korea’s US$6.4B Binance heist took 54 minutes and isn’t over appeared first on e27.

Bangkok, October 26, 2026 – SCB 10X, the disruptive technology investment arm of SCBX Group, announced its investment in Tasklet, an AI agent platform that enables businesses to automate complex workflows and delegate operational tasks to AI agents. Tasklet has raised an additional $12.8 million in funding, with participation from investors including Union Square Ventures, 359 Capital, Uncorrelated, Robinhood Ventures, Rakesh Agrawal, Patrick and John Collison, and others.

Tasklet is building an AI-powered platform designed to help businesses move beyond AI assistants toward AI agents that can take action and execute real work. The platform allows users to describe business processes in natural language, while AI agents connect with business applications, data sources, and tools to perform tasks autonomously.

Founded by Andrew Lee and Jonny Dimond, Tasklet draws on the team’s experience building Shortwave, an AI-powered email platform. Andrew previously co-founded Firebase, which was acquired by Google in 2014. The company was created around a broader vision for AI agents that can move beyond assisting with individual tasks and instead take responsibility for ongoing work across business systems.

Tasklet is designed around persistent agents that can take ongoing responsibility for work. Each agent can connect to business tools and data, respond to events and schedules, retain relevant context through memory, and operate across applications and browser environments. For enterprise use, Tasklet also provides organizational controls, permissions, guardrails and security capabilities designed to support deployment in real business workflows.

Also Read: SCB 10X Identifies Six Key Takeaways for Organizations Navigating the AI Transformation, Drawing Insights from AI-VOLUTION The Series 2026

“We see a significant long-term opportunity for AI agents to move beyond individual tasks and support ongoing business workflows. Andrew’s experience building Firebase, together with the team’s work on Shortwave, brings valuable platform-building perspective to this emerging space. We’re pleased to support Tasklet as the team develops the connectivity, collaboration, and controls that businesses will need to adopt AI agents at scale” said Pailin (Pie) Vichakul, Chief Investment Officer at SCB 10X.

“We believe the next phase of AI for businesses is not just about helping people work faster, but enabling companies to delegate real work and ongoing responsibilities to AI agents. That requires more than intelligence alone. Agents need the context, connectivity, collaboration, security and controls to operate reliably across real business workflows. We’re excited to have SCB 10X as an investor who shares our conviction in this shift and sees the potential for AI agents to become a core part of how enterprises operate.” said Andrew Lee, Co-founder and CEO at Tasklet.

For enterprises looking to put AI agents to work, Tasklet offers a practical starting point: identify a real workflow, connect an agent to the tools and context it needs, and expand from there. Agents can take on recurring or event-driven responsibilities while organizations retain visibility, permissions and control over how they operate. This enables teams to move from isolated AI use cases toward shared agent workflows that can become part of day-to-day business operations. Enterprises can learn more and explore Tasklet at tasklet.ai.

 

About Tasklet

Tasklet is an AI agent platform for automating business workflows. Founded by Andrew Lee and Jonny Dimond, Tasklet enables businesses to create AI agents that connect to their tools and data, reason through complex tasks, and execute workflows autonomously. The platform supports thousands of integrations, including APIs, MCP servers, and cloud-based browser automation, enabling AI agents to perform real operational work across business systems.

For more information, visit https://tasklet.ai/

 

About SCB 10X

SCB 10X is the disruptive technology investment arm of SCBX Group. With an investment track record since 2016, SCB 10X has deployed over USD 500 million globally into startups in AI, blockchain, and fintech. SCB 10X has backed exceptional companies such as Together AI, Pagaya, Ripple, Fireblocks, Anchorage Digital

Beyond capital, SCB 10X partners with our portfolio founders to test, grow and scale their solutions through SCBX’s network, unlocking commercial opportunities into Thailand and Southeast Asia. Mandated as the group’s speedboat, we discover and ship state-of-the-art technologies and solutions into SCBX group.

For more information, please visit https://scb10x.com/

The post SCB 10X Invests in Tasklet, AI Agent Platform for Business Workflow Automation, in Its Latest Funding Round appeared first on e27.

Singapore’s manufacturers are in their most buoyant mood in recent memory, and they largely have one customer category to thank: the AI data centre.

A net weighted balance of 24 per cent of manufacturers expect business conditions to improve between October 2026 and March 2027, the Singapore Economic Development Board (EDB) said on Monday. A weighted 29 per cent of firms project better conditions compared to the third quarter, against just 5 per cent bracing for a weaker stretch.

Also Read: AI boom pushes chipmakers to expand across Asia Pacific, FedEx report finds

The “net weighted balance” is simply the share of upbeat responses minus the share of downbeat ones, weighted by firm size. By that yardstick, the latest reading is a sharp jump. The previous survey, covering July to December 2026, put the figure at 12 per cent, while the one before that, covering April to September, came in at 17 per cent.

In other words, sentiment has doubled in a single quarter, even as geopolitical tensions continue to cloud the global outlook.

Chip-equipment makers lead the charge

The precision engineering cluster is the most upbeat of the lot, buoyed by semiconductor equipment firms that expect global AI-related investment to keep expanding. That tracks with the capital flowing into the city-state’s tool-making base, including Applied Materials’ US$500 million Singapore expansion earlier this year.

The electronics cluster is not far behind, pointing to robust orders for semiconductors used in AI applications. Its semiconductors and computer peripherals & data storage segments expect to ramp up production to serve what EDB describes as buoyant demand from AI data centres.

On output, a net weighted 34 per cent of manufacturers expect to produce more in the fourth quarter than in the third. Precision engineering firms foresee higher output of semiconductor equipment, bonding wires (the ultra-fine metal threads that connect a chip to its package) and metal precision components that feed the wider electronics supply chain.

None of this is happening in a vacuum. Singapore has been trying to anchor itself more deeply in the AI hardware stack, with initiatives such as the A*STAR- and EDB-backed SG Semiconductor, as the region jostles for a slice of AI’s bottleneck economy.

Beyond chips, the picture gets patchy

Step away from silicon, and the mood is more mixed. The biomedical manufacturing cluster expects higher output of active pharmaceutical ingredients and life science instruments. Within general manufacturing, every segment except printing projects higher output, with food, beverages & tobacco firms anticipating stronger export demand for milk and beverage products.

Also Read: Southeast Asia’s chip-hub ambition is colliding with its chip-smuggling problem

The transport engineering cluster expects broadly similar business conditions, citing rising costs of materials, fuel and freight, as well as overseas competition. Its output outlook is split: aerospace and land segments expect more activity, while marine & offshore engineering anticipates lower production of oil and gas-field equipment as geopolitical uncertainty weighs on orders.

The chemicals cluster is the gloomiest, with petroleum firms worried about persistent uncertainty in crude oil feedstock supply.

Read together, the survey tells a story of a two-speed sector. The clusters tied to AI infrastructure are carrying the headline number; the rest are treading water or worse. That is a comfortable position while hyperscalers keep pouring money into data centres, but it also means Singapore’s manufacturing optimism is increasingly a leveraged bet on a single spending cycle. Should AI capex cool, the very clusters lifting the index would be the ones dragging it down.

Optimism isn’t translating into a hiring spree

For workers, the cheer is more muted. A weighted 72 per cent of firms expect employment in the fourth quarter to stay broadly similar to the third. All clusters except biomedical manufacturing and transport engineering project a larger workforce, but the overwhelming majority are holding headcount steady.

That gap between sentiment and hiring is worth noting. Semiconductor and equipment plants are capital-intensive and heavily automated, so a surge in orders does not automatically mean a surge in jobs. Higher output, flat payrolls: that is the shape of advanced manufacturing growth in a high-cost economy.

Exporters still face a hostile world

The global backdrop remains unforgiving. A weighted 23 per cent of firms anticipate difficulties securing export orders in the fourth quarter. The top two culprits are price competition from overseas rivals and political or economic conditions abroad, including geopolitical developments, tariffs and inflation. Those concerns echo earlier findings that Singapore interpreneurs are the most cautious on overseas expansion as tariff and supply risks bite.

The competition is also local. Neighbours are chasing the same AI-driven investment: Malaysia has been selling more chips without necessarily winning more of the market, while Thailand is targeting a US$80 billion semiconductor pushto move beyond assembly.

Also Read: 15 Southeast Asian semiconductor startups moving beyond assembly

Singapore’s pitch rests on precision, reliability and higher-value work rather than cost, and part of what is unfolding is a quiet redistribution of manufacturing FDI across Southeast Asia. For now, the AI boom is giving Singapore’s factory floor its strongest tailwind in years. The harder question is what keeps the engines running once that wind changes direction.

The post Singapore’s factories are riding the AI wave, but the optimism rests on a narrow base appeared first on e27.

One photograph in a recent Nikkei Asia report says more about Southeast Asia’s tech economy than most investor decks I have read this year. It shows Myanmar refugees working on a data centre construction site in Chonburi, eastern Thailand.

Sit with that for a moment. People who spent years, sometimes decades, in border camps without the legal right to earn a baht are now pouring concrete for the server halls that will train and run the region’s AI models. The cloud, it turns out, still stands on very human shoulders.

Also Read: AI literacy in Thailand is reaching office workers, not the 63% in informal employment

Thailand did not arrive here out of generosity. It arrived here because it ran out of workers. And the way it handles that shortage is a test, not just of its human rights record, but of whether the region’s startup ecosystem is building for the economy it actually has, or the one it pitches.

A labour model held together with border passes

The trigger was the deadly Thailand-Cambodia border conflict that erupted in July 2025. Cambodian workers began heading home in their hundreds of thousands. A Thai Labour Ministry analysis later cited a Cambodian estimate of 780,000 returns, including undocumented migrants and family members, according to the South China Morning Post. The exact figure is contested, but nobody disputes the hole it left in construction, farming and fisheries.

The bigger story is how exposed Thailand was in the first place. In one recent Department of Employment count, Myanmar nationals made up 2.27 million of the 3.14 million legal workers Thailand drew from nine Southeast Asian countries. Add an ageing population and a low birth rate, and you have an economy whose factories, building sites and fishing boats run on neighbours who can be sent home, or scared home, by a single border flare-up.

Bangkok’s response has been a scramble: holding on to Myanmar workers already in the country, a pilot scheme to bring in Sri Lankans and, most consequentially, opening the refugee camps.

A historic door, opened a crack

Credit where it is due. In October 2025, Thailand let long-staying refugees in its nine border camps work legally outside them for the first time. Some of these camps date back to the 1980s. The UN refugee agency called the move a turning point, and it was. It came just as US aid cuts forced groups such as The Border Consortium to halt most assistance, leaving more than 100,000 camp residents short of food and healthcare.

Now look at the numbers. Roughly 42,000 people of working age were deemed eligible. One year in, the Thai government says 6,800 refugees have joined the labour force, according to Deutsche Welle. That is about one in six.

Also Read: Singapore’s robotics dominance is a warning sign dressed up as good news

The reasons are not mysterious. Rights groups report that many participants receive no contract, and that conditions and pay are worse than promised. Some refugees repay employers through wage deductions for medical checks, permits and transport, which means they start work already in debt.

If that sounds familiar, it should. Debt-financed recruitment is the oldest trap in Southeast Asia’s migrant labour system. Thailand has opened a historic door, then left the old toll booth standing in front of it.

Robots are not coming to the longan farm

The tech industry’s instinctive answer to a labour shortage is to automate it. Some Thai economists even urged the government to adopt new technologies to cut reliance on labour-intensive work after the Cambodian exodus. Physical AI is the hottest pitch in the region, and plenty of decks doing the rounds promise robots for warehouses, construction and agriculture.

Some of that will happen, slowly. But be honest about the timeline and the economics. The jobs the Cambodians left (the so-called 3D jobs: dirty, difficult and dangerous) sit on longan farms in Chanthaburi, on fishing boats and on scaffolding. These are messy, unstructured environments, run by small employers on thin margins. They will not be buying humanoids at robotaxi prices any time this decade.

The automation pitch also carries a quiet assumption: that the people doing this work are a temporary inconvenience to be engineered away. Thailand has more than 3 million documented migrant workers and, by most estimates, a large undocumented population on top. They are not a gap to be closed. They are the economy.

The unglamorous stack that would actually help

Here is the contrarian bet: the bigger opportunity is not replacing these workers but formalising them. Thailand’s refugee scheme is failing at exactly the points where good software works best.

Start with identity. Refugees began receiving non-citizen Thai ID cards in June, a step away from statelessness. One card can unlock a bank account, a phone contract and a wage. Pair that with e-KYC built for people whose documents come from a camp, not a junta, and you remove the first wall.

Also Read: Circle to acquire Tazapay for US$400M as USDC push moves into cross-border payments

Then fix the debt. Recruitment fees and deductions survive because they are invisible. Digital contracts in Burmese and Karen, earned-wage access and payroll that itemises every deduction would make the toll booth visible, and visible fees are much harder to inflate.

Then remittances. The informal hundi network still moves much of the money from Thailand to Myanmar, and UNESCAP has noted that strict KYC rules push migrants towards it. Formal rails already exist: DeeMoney, TrueMoney and 2C2P’s Wave app all send money into Myanmar wallets. What they lack is reach among workers that banks still treat as a compliance risk.

Finally, worker voice. Issara Institute’s Golden Dreams app already lets migrants rate employers and recruiters in Burmese and Khmer, a Tripadvisor for bosses. That model deserves venture-scale ambition, not grant-scale budgets.

None of this is a moonshot. It is fintech and HR tech, the two categories Southeast Asian investors claim to understand best, pointed at the customers they have mostly ignored.

Dignity is also a business model

The refugees now working legally in Thailand earn roughly 3,000 baht (about US$90) a week on some farms, a life-changing sum for families who lived on rations. They want to send money home, build credit and change jobs without paying a broker. That is a customer base with clear needs and no good products.

Also Read: The gold rush nobody is talking about: AI agents and the workers our economy left behind

The lesson travels. Malaysia, Singapore and Thailand all run on migrant labour, and each has its own version of the debt trap. The region’s founders love to talk about the unbanked. Very few of them build for the workers pouring the concrete under their data centres.

Thailand has shown that a government can change a 40-year-old policy when the economics force its hand. The question now is whether the startup ecosystem will help that policy work, or wait for the robots.

The post Thailand’s labour crisis and the refugees building its AI future appeared first on e27.

Banks perform a useful trick. They let customers withdraw money from checking and savings accounts at short notice, while using much of that funding to support mortgages and business loans that may not be repaid for years.

The system works because customers do not all ask for their money back at the same time. Most balances are remarkably stable. They move slowly enough that banks can treat a large part of their deposits as dependable funding, even though customers are legally free to withdraw them.

AI agents could make that assumption much less reliable.

The obvious concern is a faster bank run. Apollo economist Torsten Slok recently pointed to an average annual interest rate of 0.1 per cent on US checking-account balances while selected alternatives advertised yields of 3.3 to 5.0 per cent. An AI agent that continually searched for a better return could move money far faster than a human customer normally would.

But a bank run may not be the most important effect. The bigger change could be that banks have to start competing much harder for money they currently receive cheaply and keep for long periods.

The customer may stay while the money leaves

People do not choose bank accounts on interest rates alone.

A checking account may provide payments, cash access, customer service, deposit insurance and the comfort of dealing with a familiar institution. For some customers, those benefits justify earning less interest. That does not mean the customer must keep all their money there.

An AI agent could understand that someone wants to keep US$5,000 in a familiar bank for bills, cash withdrawals and everyday payments, but does not need another US$50,000 sitting in the same account earning very little. The agent could leave the service relationship intact while moving the surplus money somewhere that pays more.

This is important because the bank could keep the customer but lose much of the funding that made the customer profitable.

Large banks benefit from bundling many services together. Yet several parts of that bundle can already be obtained elsewhere. Smaller insured banks can provide the same standard US deposit-insurance protection within the relevant limits. Shared ATM networks can provide cash access without every bank building its own network. Customer service is increasingly automated.

As AI gets cheaper, agents could make it easier to compare the whole package rather than simply compare interest rates. They could ask: How much cash access does this customer actually need? Do they care about branches? What fees would moving money trigger? Is the alternative account insured? How much money should remain instantly available?

Automatic cash management is not new. Wealthfront introduced Autopilot in 2020, allowing customers to move excess cash according to rules they had already set. The difference with a more capable agent is that it could potentially choose between providers as well as execute transfers. The open question is whether that increase in capability is large enough to change consumer behaviour at scale.

Also Read: How a 39-store retail chain in Laos moved from Excel to AI agents

There is already evidence that greater attention changes bank pricing. Research from the Bank for International Settlements finds that US digital banks tend to pay higher deposit rates and adjust them more strongly when central-bank rates change.

Agents could make many more customers behave like highly attentive customers.

Banks would have to start paying for stability

That creates two separate problems for banks.

First, deposits become more expensive. A bank may have to raise its savings rates to stop money leaving.

Second, even after matching a competitor’s rate, the bank has no guarantee that the money will stay. If another bank improves its offer tomorrow, the customer’s agent can move again.

For banks that need reliable funding over longer periods, fixed-term deposits therefore become more valuable. Instead of relying on customers simply leaving money alone, the bank pays them for a contractual commitment to keep it there.

Interestingly, agents could help on this side too. An agent might calculate that a household will not need US$20,000 for the next year and place it into a fixed-term account. So the same technology that makes instant-access deposits less stable could increase the supply of deposits committed for longer periods.

The important change is that banks would increasingly have to pay explicitly for stability rather than receive part of it through customer inertia.

There is also a more extreme risk. Silicon Valley Bank lost US$42 billion in a single business day during its 2023 crisis. That was a confidence-driven run, not an AI event, but it showed how quickly deposits can already move.

If millions of agents use similar decision rules, withdrawals could become more synchronised. A small pricing problem could become a liquidity problem if a bank suddenly had to replace large amounts of funding at much higher cost.

Who actually pays when deposit rates rise?

Initially, the bank does.

If a bank raises the interest paid to depositors, its costs rise. It then has four broad choices: charge borrowers more, reduce its operating costs, accept lower profits, or make fewer loans. In practice, all four could happen at once.

This is why higher returns for savers do not automatically mean equally higher mortgage and business-loan rates.

Banks spend a great deal of money operating branches, call centres, compliance teams, administrative processes and large technology systems. If AI reduces those costs, banks could pay depositors more while spending less elsewhere.

Borrowers may also become more price-sensitive. Their agents could compare loan offers, prepare applications and identify when refinancing makes financial sense. That would make it harder for a bank simply to pass every increase in its funding costs to borrowers.

Also Read: Who referees the AI agents? Enterprises aren’t sure, new report shows

So the extra return paid to savers could come from several places. Borrowers may pay somewhat more. Bank profits may fall. Operating costs may fall. Some loans that no longer make economic sense may disappear.

The result depends heavily on competition.

A lean bank might pay depositors more, charge borrowers no more than today, and remain profitable because its operating costs are much lower. A bank carrying thousands of employees, expensive branches and legacy systems would have a much harder time doing the same.

This does not create specialist finance. It changes the economics of it

Specialist lenders and specialist funding providers already exist at scale.

US finance companies owned or managed about US$1.95 trillion of loans and leases in July 2026. In the UK, Hampshire Trust Bank already takes deposits while financing independent lending companies. Nonbank mortgage companies commonly borrow from banks, originate mortgages and then sell those loans to investors.

AI agents would not invent this structure. What they could change is its cost and competitive reach.

A specialist lender could automate more of its administration, reach customers more cheaply and operate profitably in smaller niches. At the same time, customers’ agents could make those specialists easier to find.

But the large banks would not simply disappear. Specialist lenders often depend on banks for funding. If bank funding becomes more expensive, some of that cost flows through to them too. Large banks also retain advantages in capital, regulation, liquidity and access to emergency funding.

Also Read: AI agents broke into 100 firms for pocket change. SEA’s online shops fit the target profile

So the likely result is not a clean transfer of business from large banks to thousands of tiny replacements. It is more likely to be a squeeze on the margins throughout the system.

Savers may capture more of the available return. Borrowers may gain from better comparison. Banks may retain less income between the two. Some of the value may move to specialist lenders, funding providers or the agent platforms themselves.

Fractional-reserve banking probably survives, but in a different form

AI agents do not automatically end fractional-reserve banking.

Banks can respond by paying more for deposits, attracting more fixed-term money, holding more liquid assets and reducing their own operating costs.

The deeper change would occur only if long-term lending became much less dependent on deposits that customers can withdraw immediately and much more dependent on money that savers and investors have committed for longer periods.

That would be more than a change in which company provides the mortgage or savings account. It would change the type of money supporting the lending.

The first test is much simpler. Once reliable agents can move money easily between banks, do large banks continue to hold huge amounts of cheap, stable deposits?

If they do, the existing model is more durable than this thesis suggests.

If they do not, one of the oldest economic advantages in banking begins to disappear.

Read “Abundance Economics: The Economic Mechanisms of the AI Age” here, or find more on the research framework behind it here.

—

The post Could AI agents break fractional-reserve banking? appeared first on e27.