---
title: "Will AI replace underwriters? What the documented record shows in 2026"
description: "Will AI replace underwriters? AI already underwrites at scale, but the documented record shows the liability, and most of the job, staying human."
url: "https://theinternetninja.com/blog/will-ai-replace-underwriters-what-the-record-shows-2026/"
updated: "2026-09-04"
---

# Will AI replace underwriters? What the documented record shows in 2026

By Maneesh Behera, Head of AI Deployment. Published 2026-09-04. The Internet Ninja, analysis.

> AI already makes underwriting calls at scale. The record shows the accountability, and most of the job, staying with a person.

## Case files this argument rests on

- [New York's financial regulator investigated the Apple Card algorithm for gender bias and found no unlawful discrimination](https://theinternetninja.com/stories/apple-card-goldman-sachs-nydfs-no-unlawful-discrimination-fair-lending-2021/)
- [Klarna's AI customer-service assistant: the 2024 numbers and the 2025 walk-back](https://theinternetninja.com/stories/klarna-ai-assistant-customer-service/)
- [Octopus Energy's 'Magic Ink': AI that drafts support emails, with no job cuts](https://theinternetninja.com/stories/octopus-energy-kraken-ai-support/)

If you underwrite loans or policies for a living, the pitch decks have already
written your obituary. Every vendor demo shows a model reading an application
and returning a decision in seconds, and the implied headcount is zero.

The record is more specific than the pitch. It shows AI already making
underwriting calls at scale, and it shows who paid when those calls went wrong.
Those are two different questions, and the second one is the job.

Start with the scale. New York's financial regulator analysed the underwriting
data for "approximately 400,000 New York State applicants" for the Apple Card, a
credit product underwritten by a machine-learning model, and found no unlawful
discrimination ([source](https://www.dfs.ny.gov/reports_and_publications/press_releases/pr202103231)).
A model underwrote nearly half a million applications. The underwriter as a
decision-maker on each file is already, in that product, gone.

## What AI underwriting means

AI underwriting is the use of algorithmic or machine-learning models to decide a
credit or insurance application: who is eligible, on what terms, at what price.
It is a decision system, not a drafting aid, which is why it carries legal
weight a chatbot does not.

## Will AI replace insurance underwriters, or move the work?

The honest answer is move, then thin, not replace. A model can score risk and
clear the clean applications, but someone still owns the edge cases and the
regulatory exposure. That ownership is the part of the job a model cannot hold,
and it is where underwriters concentrate as automation spreads.

Klarna is the cleanest warning against confusing task automation with headcount.
In 2024 it said its AI assistant was "doing the equivalent work of 700 full-time
agents" ([source](https://www.forbes.com/sites/jackkelly/2024/03/04/klarnas-ai-assistant-is-doing-the-job-of-700-workers-company-says/)).
That figure is a modeled workload equivalence, not 700 people removed.

By 2025 Klarna had reversed the posture. Its CEO said the cost-first push
produced "lower quality," and the company moved to recruit human agents again
([source](https://www.customerexperiencedive.com/news/klarna-reinvests-human-talent-ai-customer-service-buy-now-pay-later/747586/)).
That is customer service, not underwriting, but the pattern transfers: judgment
work automated on cost alone tends to come back.

## Will AI replace mortgage underwriters when the model gets it wrong?

No, because the model does not carry the liability. When an algorithmic
underwriting system harms applicants, the lender answers for it, and that answer
is expensive enough to keep a human accountable for the decision.

Massachusetts settled with the student-loan lender Earnest for "$2.5 million"
over AI underwriting models that its Attorney General said could disparately harm
"Black, Hispanic, and non-citizen applicants" ([source](https://www.mass.gov/news/ag-campbell-announces-25-million-settlement-with-student-loan-lender-for-unlawful-practices-through-ai-use-other-consumer-protection-violations)).
Independent trade coverage reports the same figure and effect, including a
"knockout rule" that auto-denied applicants without a green card
([source](https://bankingjournal.aba.com/2025/08/mass-ag-reaches-settlement-with-earnest-operations-for-2-5m-over-ai-lending-bias/)).
These are the Attorney General's allegations, resolved by settlement rather than
a court finding.

The lesson for the desk is direct. Automate the scoring and you still need a
person who can be held responsible for the rule the model applied, because a
regulator will look for one.

## The proof

TIN verified the Apple Card case against the New York regulator's own record. It
shows AI underwriting working at scale and surviving a fair-lending review, with
the lender inside the regulator's jurisdiction: read
[the Apple Card fair-lending investigation](https://theinternetninja.com/stories/apple-card-goldman-sachs-nydfs-no-unlawful-discrimination-fair-lending-2021/).

TIN verified the Klarna case the same way. It is the documented reversal of a
cost-first automation push, useful here because underwriting is the same kind of
judgment work Klarna tried to fully automate and then rehired for:
[Klarna's AI assistant and the 2025 walk-back](https://theinternetninja.com/stories/klarna-ai-assistant-customer-service/).

And TIN verified Octopus Energy's deployment, where the AI drafts and a human
reviews and sends every reply, "with a human agent reviewing and sending each
one," and no layoffs followed
([source](https://www.techuk.org/resource/case-study-kraken-tech-s-generative-ai-tool-for-customer-service.html)).
Its CEO called the tool "the work of 250 people," a workload estimate, not a cut
([source](https://sifted.eu/articles/octopus-energy-ceo-chatgpt-news)). That
human-in-the-loop shape is the one underwriting is most likely to take:
[Octopus Energy's Magic Ink](https://theinternetninja.com/stories/octopus-energy-kraken-ai-support/).

## What each case actually settled

| Case | What the AI did | Who was accountable | What the record shows |
|---|---|---|---|
| Apple Card (Goldman Sachs) | Underwrote ~400,000 credit applications | The lender, under the regulator's jurisdiction | Model cleared of unlawful discrimination; lender still answerable |
| Earnest (Mass. AG) | Underwrote student loans with a scoring and knockout rule | The lender, who paid $2.5M | Automation did not move the liability |
| Klarna | Handled customer-service work of "700 agents" | The company, which later rehired humans | Cost-first full automation walked back |
| Octopus Energy | Drafted support replies, human sent each | A human agent on every reply | Task automated, role and headcount kept |

## The bottom line

Underwriters who expect AI to do nothing are wrong, and so are the decks that
expect it to do everything. The task of scoring a clean file is already leaving
the desk. What is not leaving is the accountability for a bad decision, because
regulators fine the lender, not the model, and that fine is what keeps a
qualified human on the file. Automation reassigns the work. It does not reassign
the blame, and roles reorganise around the part a model cannot own.

## Related case files

Insurance has a parallel case in [Pennsylvania's agreement with GEICO over an AI-initiated auto insurance policy cancellation process](https://theinternetninja.com/stories/pennsylvania-ag-geico-ai-underwriting-auto-insurance-cancellation-settlement-2026/): a new Philadelphia policyholder's policy was cancelled without her realizing it, and GEICO, without admitting any violation of the law, agreed to follow the Pennsylvania Insurance Department's guidance on the use of AI by insurers.

## Sources

1. New York State Department of Financial Services, "DFS Issues Findings on the Apple Card and Its Underwriter Goldman Sachs Bank," 2021-03-23. https://www.dfs.ny.gov/reports_and_publications/press_releases/pr202103231
2. Massachusetts Attorney General's Office, "AG Campbell Announces $2.5 Million Settlement With Student Loan Lender For Unlawful Practices Through AI Use," 2025-07-10. https://www.mass.gov/news/ag-campbell-announces-25-million-settlement-with-student-loan-lender-for-unlawful-practices-through-ai-use-other-consumer-protection-violations
3. ABA Banking Journal, "Mass. AG reaches settlement with student loan firm for $2.5M over AI lending bias," 2025-08-01. https://bankingjournal.aba.com/2025/08/mass-ag-reaches-settlement-with-earnest-operations-for-2-5m-over-ai-lending-bias/
4. Forbes, "Klarna's AI Assistant Is Doing The Job Of 700 Workers, Company Says," 2024-03-04. https://www.forbes.com/sites/jackkelly/2024/03/04/klarnas-ai-assistant-is-doing-the-job-of-700-workers-company-says/
5. Customer Experience Dive, "Klarna changes its AI tune and again recruits humans for customer service," 2025-05-14. https://www.customerexperiencedive.com/news/klarna-reinvests-human-talent-ai-customer-service-buy-now-pay-later/747586/
6. Sifted, "Octopus Energy CEO: AI is doing the work of 250 people," 2023-05-11. https://sifted.eu/articles/octopus-energy-ceo-chatgpt-news
7. techUK, "Case study: Kraken Tech's generative AI tool for customer service," 2023. https://www.techuk.org/resource/case-study-kraken-tech-s-generative-ai-tool-for-customer-service.html

## Questions

**Will AI replace underwriters?**

AI will replace parts of underwriting, not the underwriter. Algorithmic models already decide eligibility, terms and pricing at scale, but the documented record shows the lender, not the model, carrying the legal accountability, so the role shifts toward supervision and exceptions rather than disappearing.

**Will AI replace insurance underwriters?**

Insurance underwriters face the same split as credit underwriters. A model can score and triage risk, but a person still owns the judgment calls and the regulatory liability, so the near-term change is a smaller team doing higher-stakes review, not an empty desk.

**Will AI replace mortgage underwriters?**

Mortgage underwriting is heavily automated already, yet fair-lending law holds the lender responsible for what the model does. Earnest paid $2.5 million over an algorithmic underwriting rule, which is why lenders keep human underwriters accountable for the decision rather than removing them.

This post is analysis, not proof. The evidence lives in the case files: https://theinternetninja.com/stories/