Carrington Labs vs Upstart (2026)
The decision is whether your institution or your vendor answers to the regulator. Carrington Labs builds a model from your own borrowers and hands it over, and it names no regulator, statute or lending rule in any market, so every supervisory question about that model lands on you. Upstart is a supervised entity in its own right, holding 248 state lending licences, a published mortgage licensing identifier and a national bank charter application, and its model was examined by the Consumer Financial Protection Bureau under a no action letter that tested whether it produced greater disparities than a traditional model by race, ethnicity, sex or age. That evidence is stronger than anything Carrington publishes and it is also perishable: the Bureau terminated the letter in 2022, and a fair lending monitorship by the law firm Relman Colfax found approval disparities in first quarter 2022 data. Both document six of the nine regulatory axes in the AI FinTech Index and neither publishes a security attestation or a hosting region. Buy Carrington Labs to keep the model and the accountability. Buy Upstart to move both.
- You have borrower data and a decision engine but no modelling team. Carrington Labs states that is precisely the gap it fills, builds the model from your own borrowers and repayment outcomes, delivers it into systems already running, and pilots a tailored model in days with onboarding in weeks.
- The lending decision has to stay yours. The company is explicit that it does not replace origination, decisioning or servicing, supplying signals that sit alongside your bureau data, policy rules and existing scorecards, with a second look pattern for testing cash flow data on borderline applications without rebuilding anything.
- You want the score decomposed rather than delivered whole. The Cashflow Score breaks into five named behavioural categories covering velocity, liquidity, stability, leverage and resilience, with model features mapping directly to adverse action reasons so a decline can be explained by your own credit officer.
- You want the lending programme, not the components. More than 90 percent of loans complete fully automated across 100 plus bank and credit union partners, spanning personal, auto, home equity and small dollar relief lending under one implementation, one compliance relationship and one account team.
- Your fair lending committee wants an outcome test rather than a process claim. Upstart's model was examined under a Consumer Financial Protection Bureau no action letter that tested whether it produced greater disparities than a traditional model with respect to race, ethnicity, sex or age.
- You would rather your counterparty be supervised than merely compliant. 248 state lending licences, a published mortgage licensing identifier, a national bank charter application and a Federal Register submission on how partners should oversee the relationship make this a regulated entity in its own right.
This comparison is published by AI FinTech Index, an independent research platform that publishes independent ratings of AI vendors for financial services. Carrington Labs and Upstart are each graded against the same capability taxonomy, from each vendor's own public materials and the regulatory record, under the AI FinTech Index verification standard. No vendor pays for placement, and no vendor has reviewed this page. How this evidence is graded
Plain facts
| Carrington Labs | Upstart | |
|---|---|---|
| Primary category | Credit Decisioning & Underwriting | Credit Decisioning & Underwriting |
| Founded | 2023 | 2012 |
| Headquarters | Sydney, New South Wales, Australia | San Mateo, California, United States |
| Website | www.carringtonlabs.com | www.upstart.com |
Side by Side
| Axis | C Carrington Labs |
U Upstart |
|---|---|---|
| AI Centrality | ||
| Autonomy and Oversight Model | ||
| Model Risk Management and Transparency | ||
| Operational and Outcome Evidence | ||
| AI Safety and Data Stewardship | ||
| GLBA and Data Privacy Posture | ||
| Security Certifications and Trust Center | ||
| Regulatory Status and Licensure | ||
| AI Governance and Bias Disclosure | ||
| AI Liability and Recourse | ||
| Model Supply Chain Disclosure | ||
| Core Systems and Integration Depth | ||
| Deployment Model and Data Residency | ||
| Commercial Transparency | ||
| Institution and Segment Coverage |
The short version of each
Carrington Labs
Carrington Labs builds cash flow underwriting models and credit risk analytics for banks and non bank lenders across consumer and small business lending, as a business of an Australian listed group, constructing each model from the lender's own borrowers and repayment outcomes and delivering it into origination, decisioning and servicing systems already running. The AI FinTech Index grades it B on model risk management and transparency, B on regulatory status, B on GLBA and data privacy posture, B on governance and bias, B on autonomy and oversight and B on model supply chain, documenting six of the nine regulatory axes the index tracks against an index average of 2.93 across 489 vendors. Its Cashflow Score runs solely on customer permissioned bank transaction data and decomposes into five named behavioural categories, with model features mapping to adverse action reasons. It names no regulator, statute or lending rule in any market. Liability and recourse, security certifications and deployment residency are each graded C.
Source: AI FinTech Index, 2026
Upstart
Upstart operates an artificial intelligence lending marketplace through which more than 100 banks and credit unions originate consumer credit using its underwriting models, with more than 90 percent of loans completing fully automated and the partner remaining lender of record. The AI FinTech Index grades it A on regulatory status and licensure, A on governance and bias disclosure and A on model risk management and transparency, documenting six of the nine regulatory axes the index tracks. It maintains 248 state lending licences and has applied for a national bank charter to be held in a separately regulated sister company. Its fair lending evidence originates in a Consumer Financial Protection Bureau no action letter issued in 2017 and terminated in 2022, under which the Bureau reported approval increases of 23 to 29 percent and rate reductions of 15 to 17 percent across tested segments based on Upstart's own analysis. Autonomy and oversight, safety and stewardship, security certifications and deployment residency are each graded C.
Source: AI FinTech Index, 2026
Common questions
Is Carrington Labs better than Upstart?
They document the same amount and document completely different things, which is why the choice is structural rather than a matter of degree. Both cover six of the nine regulatory axes. Carrington Labs spreads six B grades evenly across model risk, regulatory status, privacy, bias, autonomy and supply chain, and names no regulator, statute or lending rule in any market. Upstart holds three A grades on regulatory status, bias disclosure and model risk management, and three C grades on autonomy, safety and stewardship, and security. Carrington is uniformly moderate; Upstart is deep where it is deep and silent where it is silent. Buy Carrington if the model must fit a book only you have. Buy Upstart if the reason you are shopping is that building and running underwriting is itself the problem. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
Has either model been examined by a regulator for fair lending?
Upstart, and the qualification matters as much as the fact. Its model was examined under a Consumer Financial Protection Bureau no action letter issued in 2017, under which it agreed to share how it decides which loans to approve and how the model expands access for underserved populations, and the Bureau reported approval increases of 23 to 29 percent with average rate reductions of 15 to 17 percent across tested race, ethnicity and gender segments, based on Upstart's own analysis. That letter was terminated in 2022, and a fair lending monitorship by the law firm Relman Colfax found approval disparities in first quarter 2022 data. Carrington Labs publishes no fairness testing, disparity analysis or approval outcome by borrower group at all. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
How much do Carrington Labs and Upstart cost?
Neither publishes a rate card and both give you something else instead. Carrington Labs publishes implementation speed, with a tailored model piloted in days and a lender onboarded in weeks, which bounds the implementation cost without indicating the licence cost. Upstart states that it seeks to set transparent pricing of services for its bank partners, and as a listed company its revenue composition and fee structure appear in periodic filings any prospective partner can read before a sales call, including the disclosure that one customer generated roughly a third of revenue in the most recent year. Neither lets you compare a number without negotiating. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
Where does each vendor process our borrower data, and what security assurance is published?
Nothing published, on either side. Both grade C on security certifications and C on deployment and data residency in the AI FinTech Index. No attestation, certification, trust centre or enumerated framework was located for either, and neither names a hosting provider, region selection or residency commitment. Carrington Labs is based in Australia and serves lenders internationally with three countries excluded, which makes processing location a live question for any buyer outside its home market. Upstart operates in a single country, so cross border transfer is unlikely to arise, and a supervised institution outsourcing consumer credit decisioning will still be expected to tell its examiners where the processing occurs.
Which one lets us keep our own underwriting model?
Carrington Labs, by construction. It builds the model from your own borrowers and repayment outcomes and delivers it into the origination, decisioning and servicing systems you already run, explicitly not replacing them, and its use is described as lender controlled. Upstart does not offer that: the partner sets credit policy, business objectives and risk appetite and remains lender of record, but the model weighing more than 3,000 variables belongs to Upstart and retrains daily against repayment data from across the whole partner base. Control at design time is real at Upstart. Control at decision time is not described, with no referral threshold, manual review path or confidence exposure published. Graded by AI FinTech Index against the same capability axes from each vendor's own published materials, verified August 23, 2026. No vendor pays for placement.
Related comparisons
Other published head to head assessments involving these vendors or their closest peers. The full set for this category is on the Credit Decisioning & Underwriting page.
Upstart's fair lending evidence rests on analyses conducted under a Consumer Financial Protection Bureau no action letter issued in 2017, and that letter was terminated in 2022. A separate fair lending monitorship of the model conducted by the law firm Relman Colfax identified approval disparities in first quarter 2022 data while the letter was still in force.
The access to credit figures Upstart cites, approval increases of 23 to 29 percent and average rate decreases of 15 to 17 percent across tested segments, were reported by the Bureau based on Upstart's own analysis rather than independently produced.
Carrington Labs states up to 30 percent higher accuracy than traditional credit models with no validation methodology, discrimination statistic or independent testing published, and is a business of an Australian listed group whose parent also runs a consumer lending operation into whose pooled loan performance its scoring product is trained.