AI Agents in Finance: The Trust Problem Microsoft and Chainalysis Say Banks Cannot Ignore
AI agents are executing financial transactions without human sign-off. Microsoft and Chainalysis say banks are unprepared for what that means for trust.
Every layer of trust in modern finance rests on the same assumption: somewhere in the chain, a human signed off. Someone approved the loan, someone authorised the transfer, someone was legally responsible. AI agents are dismantling that assumption faster than the regulation designed to hold it together. Speaking at an event hosted by Alchemy in New York on April 28, 2026, executives from Microsoft and Chainalysis said the financial industry is approaching a point where machines execute transactions at scale, with no settled answer on whether those machines can be trusted, audited and controlled.
The accountability gap automation left open
Every previous wave of financial automation preserved a human at the critical juncture. ATMs dispensed cash, but a person set the limits. Algorithmic trading moved markets at speed, but a compliance officer reviewed the logs. The new generation of AI agents is built to execute without waiting for human approval, at a scale and speed that makes after-the-fact review the only viable option.
Bill Borden, corporate vice president of worldwide financial services at Microsoft, framed the problem precisely at the Alchemy event. The automation question, he said, was settled long ago: technology can execute a hedging strategy, manage a portfolio, settle a transaction. The harder question is whether those systems can be trusted when they act without direct human input, and whether a regulated institution can demonstrate, after the fact, that a machine followed the right policy. Microsoft is building infrastructure to make that demonstration possible, including tools that give AI agents distinct identities, defined permissions, and traceable action logs.
What crypto already knows about machines transacting at scale
Jonathan Levin, co-founder and CEO of Chainalysis, offered a concrete model for what automated finance at scale actually looks like. The crypto sector has operated agent-like financial infrastructure for years: smart contracts execute autonomously, software wallets transact without human instruction, and the volume is substantial. At the Chainalysis Links conference earlier in April 2026, Levin noted that stablecoins alone are now settling over $20 billion per day. The risk management disciplines built around that activity, including monitoring fund movements across thousands of wallets simultaneously, give the crypto sector a practical head start that traditional fintech is only now beginning to catch up with.
The convergence is already visible inside major institutions. At the same conference, executives from Citi described moving away from legacy batch processing toward round-the-clock commerce. The DTCC outlined a path to tokenise over $100 trillion in assets using compliance-aware infrastructure. Standard Chartered reported finding more risk in traditional payment rails than in onchain systems. Visa's decision to partner with WeFi to build onchain banking infrastructure, announced the same week, sits inside the same current. The institutions are not waiting for the governance rules to arrive.
The governance rules that do not exist yet
Both Borden and Levin expect multiple systems to coexist rather than one replacing another. Levin said most commerce will settle on public infrastructure within a decade. Borden described a more layered outcome, where public blockchains, private networks and traditional payment rails continue to operate alongside each other, connected by software.
The ability to trace a decision, question it, and assign responsibility for it is what that connecting layer still needs to earn, and the question of who is accountable when an AI agent gets it wrong has no settled answer yet.
Revolut building a physical store in Barcelona and Visa choosing a human with credibility to close the last mile of onchain banking point to the same instinct: the industry knows that trust does not come with the infrastructure. It has to be constructed separately, and deliberately.
Editor's note
Every piece published on The Bright Minded goes through careful verification, but mistakes can happen. If you spot an error, have additional information, or want to flag anything, write to rosalia@thebrightminded.com.
