TL;DR
The vertical AI landscape in financial services is consolidating around three concrete moves: (1) established fintech platforms like Mercury and Upstart are securing regulatory approval and banking talent to move upstream into institutional finance; (2) the White House is actively pushing federal regulators to lower barriers for fintech entry, while states like Colorado are reframing AI regulation around transparency and human review rather than blanket restrictions; and (3) AI agents are moving from proof-of-concept into production workflows across lending, claims, payroll, and commerce — but critical liability and infrastructure gaps remain unresolved. The winners will be vertical specialists who can articulate domain-specific AI value, not horizontal AI labs adapting to finance.
Institutional Fintech Is Crossing Into Banking Territory
Mercury and Upstart are no longer competing for consumer wallet share — they're embedding themselves into the banking system itself through regulatory approval, capital raises, and board appointments that signal a permanent move upmarket.
Mercury raised $200 million at a $5.2 billion valuation while simultaneously receiving conditional approval from the Office of the Comptroller of the Currency to establish Mercury Bank, N.A. — a fully chartered national bank. The milestone matters because it gives Mercury direct Zelle access, an expanded lending product suite, and deeper payment infrastructure it controls, reducing reliance on sponsor banks. Mercury now serves 300,000+ customers including one in three U.S. startups, with 73% of new customers coming from outside AI and tech. The company is building a natural-language banking interface (Mercury Command, launching late 2026) that lets founders check cash positions, set auto-transfer rules, and manage invoicing through conversational AI — all grounded in real account data.
Separately, Upstart appointed Tim Wennes, former president and CEO of Santander Holdings USA, to its board. Wennes brings 30+ years in retail banking and auto finance — precisely the categories Upstart is targeting — signaling that AI lending platforms are now recruiting senior banking operators rather than relying on Silicon Valley growth leadership. Over 100 financial institutions now use Upstart's AI models, with more than 90% of loans on its platform fully automated.
What to watch: Mercury's final OCC approval timeline and whether its IPO ambitions (CEO Immad Akhund explicitly stated intent not to sell) create a new public-market category for AI-native banking infrastructure.
Regulatory Tailwinds Are Real, But Fragmented
The White House has issued the most explicit directive to date pushing federal regulators to lower barriers for fintech, but implementation will be messy and state-level rules are diverging sharply.
On May 19, 2026, President Trump signed an Executive Order directing all federal financial regulators to identify rules that "unduly impede fintech firms" from entering partnerships, seeking charters, or obtaining licenses. The order includes a specific 120-day directive to the Federal Reserve to evaluate legal frameworks for expanding access to Reserve Bank payment accounts by non-bank financial companies — addressing a long-running fintech grievance that denial of master account access prevents fair competition in payments.
Simultaneously, Colorado passed a substantial rewrite of its AI consumer protection law (SB 26-189, effective January 1, 2027) that replaces the 2024 framework with a narrower focus on "automated decision-making technology" affecting consequential decisions like lending and insurance. The new law eliminates requirements for developer bias testing and public statements, replacing them with deployer obligations to provide plain-language explanations of adverse decisions within 30 days and consumer rights to request human review — effectively aligning Colorado's framework with existing federal financial regulatory standards (ECOA, FCRA) rather than imposing new substantive restrictions.
The divergence matters: federal regulators are being told to remove barriers while states are layering on transparency obligations. Financial institutions deploying AI will need to navigate both, but the direction is permissive rather than restrictive at the federal level.
What to watch: Whether the Federal Reserve's 120-day review on payment account access produces expanded access for non-bank fintechs, and whether other states follow Colorado's approach of aligning AI regulation with existing financial law rather than creating new AI-specific restrictions.
AI Agents Are Moving Into Production Workflows Across Three Domains
The conversation has shifted from "can AI agents work in finance" to "which specific workflows are being automated first." Lending, claims, and payroll are leading.
Anthropic launched 10 pre-built AI agents for financial services workflows including pitchbook building, closing the books, and credit memo drafting — announced at a New York event with JPMorgan CEO Jamie Dimon on stage. Goldman Sachs, Citadel, Citi, and AIG are among Anthropic's customers for Claude for Financial Services, which is now Anthropic's second-largest industry vertical by enterprise revenue after tech. Anthropic is also partnering with FIS on a $1.5 billion joint venture to sell AI tools to companies backed by private equity.
In insurance, insured.io launched "Claims AI," a virtual agent handling First Notice of Loss intake across voice and chat, integrating directly with carriers' core platforms for real-time policy data and claims submission. The company cites research suggesting AI-powered claims technologies can improve productivity by up to 80% and increase classification accuracy by 30% versus manual workflows.
In global payroll, RemotePass raised $17.4 million in Series B funding while reporting $800 million+ in cross-border payroll facilitated and 35,000+ workers across 150+ countries. The company has deployed AI agents to automate workflows across onboarding, compliance, and support, and recently launched SpendCards to embed corporate expense cards into the same platform — collapsing payroll, contractor payments, and spend into one system.
These are not experimental pilots. They are production systems handling material transaction volume.
What to watch: Whether these first-wave automations (intake, payroll, expense) extend into higher-stakes decisions like underwriting and claims adjudication, and what happens when an agent makes a materially wrong call.
Agentic Commerce Remains Blocked by Liability Architecture
JPMorgan Payments has deliberately taken a measured approach to agentic commerce despite being the largest credit card issuer and merchant acquirer in the U.S. The reason is structural, not technical: the decades-old card network liability model breaks down when a fourth party (the AI agent) is inserted.
Other barriers include merchant catalogue granularity (most lack the structured data agents need), multi-item transaction protocols (current systems only support single-item purchases), and loyalty system integration. JPMorgan expects autonomous agentic transactions to first appear on merchants' own websites using their own agents — not through third-party platforms.
What to watch: Whether any payment network or issuer formally proposes a liability allocation framework for agentic transactions, or whether this remains a theoretical problem until a high-profile agent error forces the issue.
What Surprised Us
-
The White House is not just talking about fintech — it's giving regulators specific timelines and metrics. The 90-day regulatory review and 120-day Federal Reserve assessment are hard deadlines with explicit deliverables. This is materially different from prior fintech-friendly rhetoric. If the Fed's review produces expanded payment account access, the competitive moat around incumbent banking infrastructure cracks.
-
Anthropic's financial services vertical is already larger than most people realize. It's the second-largest vertical by enterprise revenue after tech, with four marquee customers (Goldman Sachs, Citadel, Citi, AIG) and a $1.5B joint venture with FIS in motion. The OpenAI vs. Anthropic IPO race is not theoretical — both are building financial services revenue to disclose to investors.
-
Colorado's AI law rewrite is a template, not an outlier. By aligning ADMT regulation with existing financial law (ECOA, FCRA) rather than creating new AI-specific substantive restrictions, Colorado avoided the trap of regulating the technology rather than the decision. Other states will likely copy this approach, which means financial institutions won't face a 50-state patchwork of AI rules — they'll face a patchwork of transparency/human-review requirements layered on top of existing credit and insurance law.
-
Vertical domain expertise is now table stakes. insured.io's CPO explicitly stated that the Claims AI agent was "created by insurance experts who live and breathe insurance processes" — not adapted from a generic AI platform. ITC Infotech and InsureMO's partnership works because InsureMO's APIs are built in open, structured formats that AI agents can discover and compose directly. The winners won't be horizontal AI labs; they'll be specialists who can articulate why their vertical matters.
Open Threads Worth a Vote
-
OpenAI and Anthropic IPO preparations — financial services vertical revenue disclosures — Both are filing for IPO imminently. What financial-services-specific revenue numbers do they disclose, and does it signal a shift in how Wall Street values AI companies?
-
Post-Brex/Capital One acquisition: competitive dynamics in AI-native startup banking — Brex was acquired by Capital One in January 2026 for $5.15B. Mercury says it won't sell. Who's gaining share in this space, and are any other players being acquired?
-
Anthropic Mythos model: financial services deployment and regulatory response — ~50 organizations including major banks have preview access to Mythos. Track how it gets used in financial services and what safeguards regulators require.