Why Banks And Amazon Block AI Agents From Your Money

forbes.com · Feb 26, 2026 · Collected from GDELT

Full Article

Close-up of a person's hand holding a smartphone and using the Opus 4 model within the Claude app from AI company Anthropic, Lafayette, California, May 22, 2025. (Photo by Smith Collection/Gado/Getty Images)Gado via Getty ImagesFinancial institutions lose €5.7 billion annually in Europe alone when 68% of customers abandon account applications rather than upload the same identity documents they submitted to another bank last week. Retailers hemorrhage another $260 billion in abandoned shopping carts. Insurance companies watch 11% of customers delay switching providers purely to avoid paperwork.Isn’t this a perfect task for AI agents?Yet every major bank requires human verification for AI-initiated transactions. Amazon blocks AI agents from its $575 billion marketplace. OpenAI’s Instant Checkout requires human approval at every step. Things are not much better in crypto. The creator of ElizaOS, a popular open-source crypto AI agent framework, admits: “We launched a trading bot, and it was very good at losing money.” The standoff creates a three-way disconnect. Consumers expect AI agents to handle their finances autonomously. The technology itself—even sophisticated models like GPT-4 and Claude—can process financial tasks when given proper context. But the infrastructure doesn't exist to let them actually do it. Financial institutions won't grant permission. Retail giants protect their data moats. Banks cite regulatory compliance.When AI models work perfectly, they still hit a wall: no permission layer exists to bridge capable technology with willing consumers. And so users remain stuck uploading the same documents, entering the same data, proving the same identity over and over again.This is exactly the problem San Francisco based start-up Sumvin aims to solve.MORE FOR YOUThe Financial Services LockdownThe resistance starts with banking, where Know Your Customer (KYC) regulations create an impenetrable barrier to AI automation.Here's the annoying part about banks: they want your passport. Always. Then almost everyone needs proof of address. Most want a selfie where you turn your head and blink like you're auditioning for something. And you've already done this dance with three other banks, but each one acts like you're a complete stranger.Industry research shows people give up after about 18 minutes. No surprise there - on average people are managing 168 passwords for accounts that all demanded the same documents. It's an incredibly annoying world where you have to remember which email you used, which password variation, and whether you saved your data (risking a breach) or have to type it all again.The repetition never stops. Auto insurance wants your license, your car's VIN, five years of driving history. Home insurance needs to know how old your roof is, what your HVAC looks like, how far you are from a fire station. Life insurance basically wants your medical autobiography. Mortgages? Forget it - two months of hell with most people saying it's as bad as house-hunting itself.Try doing this across borders and watch the friction multiply. That W-8BEN tax form? Expires every three years, needs to be filed separately with each bank, takes seven hours according to the IRS. Bangkok Bank just froze a bunch of American accounts without warning because of FATCA rules. People couldn't access their own money for daily expenses.Retail giants like Amazon have their own reasons for blocking AI agents. The company generated $15.7 billion from advertising in Q2 2025 alone—revenue that depends on controlling what users see and when they see it."Amazon derives as much, if not more, value from the user-level data artifacts created by a transaction than from the transaction itself," explains Eric Seufert, a digital advertising analyst who has extensively studied platform economics. Vladimir Shmidt, a former Google executive who oversaw fintech portfolios points out: “When an AI agent skips straight to a purchase without browsing, comparing, or clicking through recommendations, platforms don't just lose one transaction's worth of ad revenue - they lose all the data that tells them what to show the next ten million users.”Why Financial Institutions Fear Autonomous AIAI models can already handle most financial tasks. They can parse documents, fill forms, make calculations, and follow complex instructions. But three fundamental issues keep banks and financial service providers from granting them permission:Regulatory ComplianceAnti-money laundering laws require financial institutions to verify customer identity and monitor transaction patterns. When an AI agent initiates a transfer, who's legally responsible? Current regulations assume human decision-making. Without clear frameworks for AI delegation, banks face unlimited regulatory liability.The Authentication CrisisFinancial services spend billions annually on identity verification because regulators demand it. But how does a bank verify an AI agent has legitimate authority to access accounts? How does an insurance company confirm the bot filing claims represents the actual policyholder?The average mortgage application already requires multiple forms of income verification, bank statements, tax returns. Now financial institutions must figure out how to authenticate not just humans, but their digital representatives. One mistake could trigger federal investigation.Fraud Prevention SystemsBanks' fraud detection algorithms look for unusual patterns—what AI agent behavior looks like. Rapid transactions, perfect form completion, instant decision-making all trigger security flags designed to catch bots. The systems built to protect money actively prevent AI from managing it.The Hidden Tax on FinanceWhile banks and retailers protect their systems, the friction tax compounds daily.Research shows financial applications see 68% abandonment rates, with €5.7 billion lost annually in European financial onboarding alone.11% of consumers delay switching insurance changing providers solely due to re-entry requirements. Mortgage applicants provide the same documents multiple times because they expire during the lengthy approval process. One frustrated applicant described "providing your homeowners insurance to your loan officer just for your loan processor to ask for homeowners insurance again two days later."The retail side isn't better. Cart abandonment hits 70%, with checkout flows averaging 23.48 form elements when 12 would suffice. Financial data—credit cards, banking details, billing addresses—must be re-entered at every merchant despite being identical across transactions.International financial services face bigger challenges. The EU maintains 24 official languages, 86 different ID card versions, and 181 types of residence documents—none portable between institutions. Foreign banks routinely refuse American clients entirely to avoid FATCA reporting requirements.The Crypto Reality CheckThe cryptocurrency world's experience with AI agents offers a preview of autonomous finance's challenges."The most disappointing thing about Eliza was, like, we could be the open claw, but everyone just wanted a trading agent," Shaw from ElizaOS told me. He explained the core problem: "It's really obvious that the reason that people make money in meme coins is almost entirely insider trading, and that it's not about how good is your LLM, it's like, how good is your information."Despite ElizaOS becoming a popular framework for crypto AI agents, practical applications remain limited. Shaw noted the lack of creativity: "Every person who used Eliza in crypto made like, either a launch pad or an reply guy or a trading wallet, and that's it, and nothing else."Even sophisticated AI struggles with basic financial tasks. ElizaOS now pivots to "agentic gaming"—using synthetic worlds to generate training data because real financial markets proved too complex for current AI capabilities.Shaw explained their pivot to gaming: "We're building games and experiences around this...games are really like, for us, how do we get that same data that anthropic is getting, you know, to build agents that get smarter."This retreat from real finance to synthetic worlds shows the infrastructure gap. It's not that AI can't understand financial markets—Shaw noted that on small models "they get, like, dramatically better after being fine tuned very quickly." It's that even perfect AI hits the permission wall: no bank will let it trade, no broker will accept its orders, no payment processor will approve its transactions.Enter the Delegated Finance InfrastructureInto this standoff steps a new category of companies betting they can solve the permission problem without threatening existing financial and retail systems.Sumvin, a San Francisco company that just raised over $1 million in pre-seed funding, is building what it calls a "permissioned, AI-powered delegated finance platform." Rather than trying to circumvent banks or compete with payment processors, they're building infrastructure both financial institutions and retailers need.Simon Jones, who runs Sumvin, puts it simply: "AI has reached the point where it can take structured, real-world actions — not just generate content. But for AI to act in consumer finance, it needs permission and verifiable infrastructure."Jones is a serious character, he has been behind the boom in crypto cards and has helped launch the MetaMask Card at Baanx that let 100 million crypto users spend from their wallets. Exodus bought Baanx for $175 million last November, and Sumvin is Jones’ new venture. What Sumvin does is basically remove friction from commerce - think of it as agent lubricant for financial services. You do KYC once, they store it on the Sei blockchain using cryptography. Therefore your AI agent can prove who you are without exposing all your data every single time.Here's why banks might actually care: they get compliance without the hassle. A bank can check if an AI agent has permission to move