Upstart Holdings has reached a milestone that could influence how future AI-driven financial institutions are regulated. The company has received conditional approval from the Office of the Comptroller of the Currency (OCC) to establish Upstart Bank, N.A., entity with potential to become the first nationally chartered bank built from the ground up with AI-powered underwriting.
The approval, however, is far from a green light to begin banking operations. Upstart must still obtain deposit insurance from the Federal Deposit Insurance Corporation (FDIC) and approval from the Federal Reserve to become a bank holding company. Before opening its doors, it must also satisfy the OCC’s customary conditions for a de novo national bank, including requirements covering capitalization, governance and operational readiness.
“It’s important for the public to understand that efficiency doesn’t diminish oversight,” said Annie Delgado, Upstart’s Chief Risk Officer and the proposed Chief Executive Officer of Upstart Bank, N.A. “A well-run charter process can be both timely and rigorous. We’ve been challenged extensively throughout the process, and that’s exactly what should happen when an institution is seeking the privilege of becoming a national bank.“
Those conditions may ultimately prove more significant than the charter itself. Much of the discussion around artificial intelligence in financial services has focused on whether machine-learning models can improve lending decisions. The Upstart case shifts the conversation toward regulation: what happens when a bank’s primary underwriting engine is an AI model rather than a traditional rules-based system?
What makes Upstart an AI-native bank?
Upstart is an AI-native lending infrastructure company whose primary product is machine-learning-based credit underwriting used by partner banks. Here’s how it works.
Conventional underwriting relies primarily on fixed credit rules and scorecards. Lenders evaluate factors such as credit score, income, debt-to-income ratio, employment history, and repayment history against predefined thresholds. Applicants who meet the criteria are approved, while those who fall outside them are often rejected or referred for manual review.
Upstart’s AI-native underwriting replaces much of this rule-based approach with machine learning. Instead of relying mainly on a credit score and static rules, its models analyze hundreds or thousands of data points simultaneously to estimate an applicant’s probability of repayment and expected credit risk. The AI identifies patterns that traditional scorecards may miss, allowing partner banks to price loans and make approval decisions based on predicted risk rather than fixed cutoffs.
Conventional Credit Decisions vs ML Underwriting: What’s the Difference?
Unlike conventional credit decisioning, alternative machine-learning models continuously identify statistical relationships across large datasets. Besides, many financial institutions today find it hard to rely on legacy financial data alone, increasingly using alternative credit data, i.e. rental and utility payments, BNPL credit, gig economy income, etc. That helps many consumers boost their credit trustworthiness in modern flexible fintech environment, but also simultaneously is making governance, monitoring and explainability central supervisory concerns. As a result, regulators are likely to scrutinize not only lending outcomes but also how models are validated, updated, documented and controlled throughout their lifecycle.
Does the Same Rules Apply to AI-Native Lenders as to Traditional Banks?
The OCC’s conditional approval suggests that regulators are not creating a separate framework for AI-native banks. Instead, they appear to be applying longstanding prudential expectations, such as adequate capital, effective governance and operational preparedness, to institutions whose core technology happens to rely on artificial intelligence. In that sense, the charter could become an important reference point for future fintechs seeking national bank status.
Questions around fair lending are also likely to remain central. In 2024, independent monitor Relman Colfax concluded a multi-year review of Upstart’s AI lending models. The monitor reported that it did not find quantitative evidence that variables used in the company’s models acted as close proxies for protected characteristics or produced pricing disparities. At the same time, it identified approval disparities affecting Black applicants and argued that an alternative model with fewer disparities could be viable, highlighting the continuing debate over how AI models should be evaluated under fair-lending rules.
That distinction illustrates why governance, rather than technology alone, is likely to define the next phase of AI adoption in banking. Regulators are increasingly focused not simply on whether an algorithm performs well, but on whether institutions can demonstrate that its development, deployment and oversight meet established supervisory standards.
PaySpace Magazine Global believes that if Upstart ultimately secures its remaining approvals and begins operating as a national bank, its governance framework could become an early benchmark against which future AI-powered bank charter applications are measured.
