An AI underwriting company from the start: in-house credit models sold to banks and credit unions as a marketplace, now seeking its own national bank charter built around those models.
How is Upstart using AI?
Upstart's AI strategy is its business: in-house credit models that price and approve personal, auto and home-equity loans for more than 100 banks and credit unions. Its main personal-loan model uses over 2,500 variables trained on nearly 104 million repayment events, and 91% of loans in 2025 were fully automated with no human intervention by Upstart. Co-founder Paul Gu became CEO on May 1, 2026, and the OCC conditionally approved Upstart Bank, N.A. on July 23, 2026.
Category
BNPL & lending
Founded
2012
Headquarters
Burlingame, California, United States
Licence and supervision
AI lending marketplace connecting borrowers to more than 100 partner banks and credit unions, holding state licences and registrations where required and subject to CFPB and FTC authority; OCC conditional approval (July 23, 2026) for Upstart Bank, N.A., a branchless Delaware-based national bank, with FDIC deposit-insurance and Federal Reserve bank-holding-company applications pending.
Scale
$4.2 billion of loan originations (558,014 loans) in Q2 2026, up 50% year over year (quarter to June 30, 2026)
Flagship platform
Upstart AI underwriting models
AI leadership
Paul Gu · Grant Schneider · Annie Delgado · Dave Girouard · Sanjay Datta
- Upstart's 2025 10-K says its personal-loan model expanded from 23 variables (end of 2014) to over 2,500 (end of 2025), was trained on nearly 104 million repayment events, and is one of approximately 30 proprietary machine-learning models built in-house.
- In 2025, 91% of Upstart-powered loans were fully automated with no human intervention by Upstart, compared with about 70% at its December 2020 IPO; Upstart credits verification and fraud-detection models for much of the gain.
- An internal 2025 study cited in the 10-K found Upstart's AI model approves 43% more borrowers than a traditional credit-score model with 33% lower average APRs; results Upstart supplied to the CFPB in 2019 showed 27% more approvals and 16% lower APRs.
- Upstart received the CFPB's first-ever no-action letter in September 2017 and a second, covering its AI underwriting and pricing model, in November 2020; the CFPB terminated it in June 2022 at Upstart's request after Upstart said it planned to add a significant number of new variables to the model.
- The SEC subpoenaed Upstart on November 17, 2023 over disclosures about its use of AI models and loans, which Upstart disclosed in May 2024; the subpoena is not mentioned in Upstart's 2025 10-K or Q2 2026 10-Q.
- A securities class action filed April 7, 2026 alleges Upstart overstated the accuracy of Model 22, its May 2025 underwriting model, which the complaint says overreacted to negative macroeconomic signals and cut approvals before a Q3 2025 revenue miss; Upstart has not been found liable.
- Co-founder and CTO Paul Gu, who ran product, engineering and machine learning, became CEO on May 1, 2026; Grant Schneider, one of Upstart's original machine-learning engineers, returned as CTO.
- Upstart applied in March 2026 for a national bank charter, pitched as the first bank built from the ground up on AI; the OCC granted conditional approval on July 23, 2026, with FDIC and Federal Reserve applications pending.
What is Upstart AI underwriting models?
Upstart's personal-loan underwriting model grew from 23 variables at the end of 2014 to over 2,500 at the end of 2025, covering credit experience, employment, education, bank-account transactions, cost of living and application interactions; it was trained on nearly 104 million repayment events and is one of about 30 in-house machine-learning models. In 2025, 91% of Upstart-powered loans were fully automated with no human intervention by Upstart, up from about 70% at its December 2020 IPO. More than 100 banks and credit unions use the models through Upstart's marketplace.[1][3]
What has Upstart done on AI, and when?
- Aug 4, 2026
Q2 2026: originations up 50%, back to GAAP profit[3]
Upstart reported $4.2 billion of originations (558,014 loans), revenue of $365 million (up 42%) and net income of $16.5 million, and said more than 90% of loans are fully automated.
- Jul 23, 2026
OCC conditionally approves Upstart Bank, N.A.[5][7]
The branchless, Delaware-based bank would originate consumer loans nationwide and take FDIC-insured deposits once FDIC and Federal Reserve approvals and OCC conditions are met; partner banks, credit unions and funds are expected to keep buying the vast majority of platform loans.
- May 1, 2026
- Apr 7, 2026
Securities class action over Model 22[9][10]
A complaint filed in the US District Court for the Northern District of California alleges Upstart and four executives misled investors between May 14 and November 4, 2025 about Model 22's accuracy, alleging it overreacted to negative macroeconomic signals and cut approvals; these are allegations, not findings.
- Mar 10, 2026
Upstart applies for a national bank charter[6]
Upstart said it would apply to the OCC and FDIC, and to the Federal Reserve as a bank holding company; Gu said 'the time is right to launch the first bank built from the ground up on AI'.
- Feb 10, 2026
- Nov 4, 2025
Q3 2025 revenue miss later tied to Model 22[10][9]
Upstart reported Q3 2025 revenue of $277 million against guidance of about $280 million and cut its full-year revenue guidance to about $1.035 billion; the stock fell 9.71% to $41.75 on November 5. A later lawsuit alleges the shortfall stemmed from Model 22, the underwriting model launched in May 2025, overreacting to negative macroeconomic signals.
- Nov 17, 2023
- Jun 8, 2022
No-action letter terminated at Upstart's request[15]
After Upstart told the CFPB on April 13, 2022 that it would add a significant number of new variables to its model, it asked to end the letter rather than wait for review; the CFPB terminated it, ending Upstart's special regulatory status.
- Nov 30, 2020
Second no-action letter covers the AI underwriting model[14]
The CFPB granted Upstart a new no-action letter regarding its automated model for underwriting and pricing unsecured, closed-end loans.
- Aug 6, 2019
CFPB publishes Upstart's model results[13]
Results supplied by Upstart and not replicated by the CFPB showed its model approved 27% more applicants and yielded 16% lower average APRs than a traditional model, with no disparities requiring further fair-lending analysis for minority, female and 62-and-older applicants.
- Sep 14, 2017
CFPB issues its first no-action letter, to Upstart[12][13]
The CFPB issued its first no-action letter to Upstart, which uses alternative data and machine learning in credit underwriting and pricing, conditioned on a model risk management and compliance plan and reporting on access-to-credit and fair-lending outcomes.
Where does Upstart use AI today?
| System | What it does | Status | Use case |
|---|---|---|---|
| AI personal, auto and HELOC underwriting[1] | In-house models price and approve loans in real time; the personal-loan model uses over 2,500 variables trained on nearly 104 million repayment events. | In production | Credit scoring & underwriting |
| Alternative-data credit access[1][13] | Upstart's 2025 internal study says its model approves 43% more borrowers than a credit-score model at 33% lower average APRs; under its 2017-2022 CFPB no-action letter it reported fair-lending comparisons to the Bureau. | In production | Fair lending & discrimination |
| Verification and fraud-detection models[1] | Upstart attributes the rise to 91% fully automated loans largely to more accurate verification and fraud-detection models, including detection of loan stacking. | In production | Fraud detection |
| Automated identity verification[1] | AI is applied to identity verification in the loan flow, part of the processes Upstart has automated end to end for most loans. | In production | AML / KYC |
| AI models supplied to banks and credit unions[3] | More than 100 banks and credit unions use Upstart's AI models and cloud applications to originate loans, making Upstart a third-party model provider for them. | In production | Third-party & vendor AI |
| Employee use of generative AI[1] | Upstart's 10-K says its employees and contractors use generative AI technologies in their work under internal policies, and flags risks including inaccurate or insecure generated code. | In production | Generative & agentic AI |
| AI model deployment inside a national bank[5][6] | Upstart Bank, N.A., conditionally approved by the OCC, would bring Upstart's underwriting models inside a supervised bank; its proposed CEO said Upstart aims to set the standard for modern AI model deployment in banking. | Announced | Model risk management |
Use-case links open the tracker's rules-by-use-case matrix: which documents govern each system type.
What has Upstart disclosed in numbers?
| Metric | Value | As of |
|---|---|---|
| Personal-loan model variables | 23 (end of 2014) to over 2,500 (end of 2025)[1] | Dec 31, 2025 |
| Repayment events in model training | nearly 104 million; one of approximately 30 in-house ML models[1] | Dec 31, 2025 |
| Loans fully automated | 91% in 2025, vs about 70% at the December 2020 IPO[1] | Dec 31, 2025 |
| Approval and APR vs a credit-score model | 43% more borrowers approved, 33% lower average APR (Upstart internal study, 2025)[1] | Dec 31, 2025 |
| Results reported to the CFPB | 27% more applicants approved, 16% lower average APRs vs a traditional model[13] | Aug 6, 2019 |
| Q2 2026 originations / revenue | $4.2 billion (558,014 loans, up 50% YoY); revenue $365 million (up 42% YoY)[3] | Jun 30, 2026 |
| Full-time employees | 1,405[1] | Dec 31, 2025 |
“The time is right to launch the first bank built from the ground up on AI”[6]
Paul Gu · Co-founder and CTO (incoming CEO) · Mar 10, 2026
“If approved, we look forward to working directly with the OCC, FDIC and the Fed to set the standard for modern AI model deployment within the banking system”[6]
Annie Delgado · Chief Risk Officer; proposed CEO of Upstart Bank · Mar 10, 2026
“Upstart Bank will allow us to lower the cost of lending and bring our full product offering to all 50 states, advancing our mission to radically reduce the cost and complexity of credit for all Americans.”[5]
Paul Gu · Co-founder and CEO · Jul 23, 2026
“It's important for the public to understand that efficiency doesn't diminish oversight”[5]
Annie Delgado · Chief Risk Officer; proposed CEO of Upstart Bank · Jul 23, 2026
Who runs AI at Upstart?
Paul Gu
Co-founder and Chief Executive Officer (since May 1, 2026; previously CTO over product, engineering and machine learning)[4][3]
Grant Schneider
Chief Technology Officer (returned February 2026; one of Upstart's original machine-learning engineers)[4]
Annie Delgado
Chief Risk Officer; proposed CEO of Upstart Bank, N.A.[5]
Dave Girouard
Co-founder and Executive Chairman (CEO until April 2026)[4]
Sanjay Datta
President and Chief Capital Officer[4]
Which regulators govern Upstart's AI?
| Authority | Why it matters here | Documents |
|---|---|---|
| CFPB | Upstart's 10-K says it is subject to CFPB-administered laws including ECOA/Regulation B; the CFPB granted it no-action letters on its machine-learning underwriting in 2017 and 2020 and terminated the second in 2022 when Upstart planned major model changes. | ECOA / Regulation B adverse action (15 U.S.C. 1691(d); 12 CFR 1002.9) CFPB Circular 2022-03 |
| OCC | Conditionally approved Upstart Bank, N.A. on July 23, 2026; the bank cannot open until capitalisation, governance and operational-readiness conditions are met, and would run Upstart's AI underwriting inside a national bank. | |
| FDIC | Upstart's application for FDIC deposit insurance for Upstart Bank, N.A. is pending. | |
| Federal Reserve | Upstart's application to become a bank holding company is pending with the Federal Reserve. | |
| SEC | The SEC subpoenaed Upstart in November 2023 over its disclosures about its use of AI models and loans. |
Linked authorities have a page in the AI Regulation Tracker; the others are named for completeness and not yet tracked.
What does the public record suggest about Upstart's AI strategy?
Analysis by BankingNewsAI from the sources cited on this page. Observations, not advice.
01
The CFPB relationship ended over the speed of model change
Upstart's no-action letter required notice of significant model changes before implementation; when it planned to add many new variables in 2022, it gave up the letter rather than wait for review. The model has since grown to over 2,500 variables.
02
Automation is Upstart's own metric, defined narrowly
'Fully automated' means no human involvement required by Upstart from rate request to funding for personal loans, but only to approval for auto loans and HELOCs. The 91% (2025) figure measures process automation, not model accuracy.
03
Approval-lift figures come from Upstart's own studies
The 27% (2019) and 43% (2025) approval lifts are Upstart's analyses; the CFPB said it did not replicate the 2019 results.
04
A charter moves the model from vendor to bank
Today more than 100 banks and credit unions use Upstart's models as a third party. Upstart Bank, N.A. would put the same models under direct OCC, FDIC and Federal Reserve supervision, which Upstart's risk chief frames as setting a standard for AI model deployment in banking.
05
Model recalibration is now a securities-law question
The April 2026 class action turns on how Model 22 reacted to macroeconomic signals and how that was described to investors, following the 2023 SEC subpoena on AI-model disclosures. Model behaviour and model disclosure are both on the record as litigation and supervisory risks.
06
Leadership now sits with the builder of the models
The 2026 succession put the co-founder who ran product, engineering and machine learning in the CEO seat and brought back an original machine-learning engineer as CTO.
How does Upstart use AI?
Upstart uses about 30 in-house machine-learning models to underwrite and price personal loans, auto loans and home-equity lines, verify identity and detect fraud. Its main personal-loan model uses over 2,500 variables trained on nearly 104 million repayment events, and 91% of Upstart-powered loans in 2025 were fully automated with no human intervention by Upstart. More than 100 banks and credit unions use the models.
Is Upstart's AI better than FICO?
Upstart says so, based on its own studies. Its 2025 10-K cites an internal study finding its AI model approves 43% more borrowers than a traditional credit-score model with 33% lower average APRs. Earlier results Upstart gave the CFPB in 2019 showed 27% more approvals and 16% lower APRs; the CFPB said it did not replicate those results.
What happened to Upstart's CFPB no-action letter?
The CFPB gave Upstart its first-ever no-action letter in September 2017 and a second one covering its AI underwriting and pricing model in November 2020. The 2020 letter required advance notice of significant model changes; after Upstart said in April 2022 it would add many new variables, it asked for termination and the CFPB ended the letter on June 8, 2022.
Is Upstart a bank?
Not yet. Upstart runs an AI lending marketplace whose loans are funded by more than 100 partner banks and credit unions and by institutional investors. On July 23, 2026 the OCC conditionally approved Upstart Bank, N.A., a branchless, Delaware-based national bank; FDIC deposit insurance and Federal Reserve bank-holding-company approvals were still pending, and the bank cannot open until OCC conditions are met.
Who is Upstart's CEO?
Co-founder Paul Gu became CEO on May 1, 2026, succeeding co-founder Dave Girouard, who remains executive chairman. Gu previously served as CTO over product, engineering and machine learning; Grant Schneider, one of Upstart's original machine-learning engineers, returned as CTO in February 2026.
What is Upstart Model 22 and why are investors suing?
Model 22 is the version of Upstart's AI underwriting model launched in early May 2025. A securities class action filed April 7, 2026 in the Northern District of California alleges Upstart and four executives overstated its accuracy between May 14 and November 4, 2025, claiming it overreacted to negative macroeconomic signals and reduced approvals; Upstart missed Q3 2025 revenue guidance ($277 million vs about $280 million) and the stock fell 9.71%. The claims are allegations.
Did the SEC investigate Upstart's AI?
Upstart received an SEC subpoena on November 17, 2023 about its disclosures on its use of AI models and loans, and disclosed it in May 2024, saying it was cooperating. The subpoena does not appear in Upstart's 2025 annual report or its Q2 2026 quarterly report; no outcome is described in those filings.
- Upstart Holdings Annual Report (Form 10-K) for the year ended December 31, 2025
- Upstart Holdings Quarterly Report (Form 10-Q) for the quarter ended June 30, 2026
- Upstart Announces Second Quarter 2026 Results
- Upstart Announces Leadership Evolution
- Upstart Receives Conditional Approval from the OCC to Establish Upstart Bank
- Upstart to apply for national bank charter
- Inside Upstart's journey to get a bank charter approval
- Upstart Holdings Quarterly Report (Form 10-Q) for the quarter ended March 31, 2024
- Upstart Holdings Faces Investor Lawsuit Over Alleged AI Model Misrepresentations and Stock Drop
- UPST Stockholder Alert: Shareholder Rights Law Firm Robbins LLP Reminds Investors of Class Action Filed Against Upstart Holdings, Inc.
- Fintech Upstart says it received SEC subpoena on AI model, loans
- CFPB Announces First No-Action Letter to Upstart Network
- An update on credit access and the Bureau's first No-Action Letter
- Consumer Financial Protection Bureau Issues No Action Letter to Facilitate the Use of Artificial Intelligence for Pricing and Underwriting Loans
- CFPB Issues Order to Terminate Upstart No-Action Letter
Neobank
BNPL & lending
Payments
Card network
Brokerage
Business finance & infrastructure
When Upstart moves on AI, you'll read it here first.
when a bank, fintech or regulator moves on AI, it is in the next morning's brief · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning