# Who Is Actually Applying? Joint Intent, Co-Applicants and Spousal Signatures When an AI Agent Runs the Mortgage Application

*November 14, 2026 · 7 min read · Pranay Shetty*

> Regulation B requires evidence, at the time of application, that each co-applicant intended to apply jointly, and it bars a creditor from requiring a spouse's signature on the note when the applicant qualifies alone. Both rules are old, both are among the findings examiners write most often, and both get harder when the application is a conversation instead of a form. How we capture joint intent at the point of sale, why it is a separate event from the accuracy attestation, and what the agent is not allowed to suggest.

## Who This Is For

Chief Compliance Officers and fair-lending officers at banks, credit unions and independent mortgage banks, and Directors of Mortgage Ops who own the point of sale. Also anyone replacing a static application form with an agent that talks to borrowers, because this is one of the places where a conversation is easier to get wrong than a form.

## An Old Rule That Keeps Showing Up in Exams

Regulation B has prohibited marital status discrimination since the 1970s, and two of its most specific rules sit in [12 CFR 1002.7(d)](https://www.consumerfinance.gov/rules-policy/regulations/1002/7/). The first says a creditor may not require the signature of an applicant's spouse or another person on a credit instrument if the applicant qualifies under the creditor's standards on their own. The second is the joint intent rule. Official comment 7(d)(1)-3 says "a person's intent to be a joint applicant must be evidenced at the time of application," that "signatures on a promissory note may not be used to show intent to apply for joint credit," and that "the method used to establish intent must be distinct from the means used by individuals to affirm the accuracy of information." The comment then gives the safe harbor: "Signatures or initials on a credit application affirming applicants' intent to apply for joint credit may be used to establish intent." The [commentary to section 7](https://www.consumerfinance.gov/rules-policy/regulations/1002/interp-7/) carries all of it, and Appendix B to Regulation B has model forms with a line built for this purpose.

Compliance writers have called missing joint intent evidence one of the most frequent Regulation B problems for years, and the Federal Reserve's [Consumer Compliance Outlook](https://consumercomplianceoutlook.org/2008/fourth-quarter/q4_02/) devoted an article to marital status compliance, joint intent and spousal signatures among it, because examiners kept finding the same problems. The reason it persists is that the violation is easy to commit without noticing. A loan officer takes an application from a married couple, both names go on the 1003, both sign everything at closing, and nobody recorded that the second person chose to be an applicant rather than being added because the loan officer assumed they would be.

In mortgage, the stakes are higher than a technical finding. A spouse who signs the note is personally liable for the debt. If they did not intend to apply, the lender has made a person liable on a loan they did not ask for, and regulators have required creditors to release people from liability they were made to take on in violation of 1002.7(d).

## The Form Already Solved This, on Paper

The redesigned Uniform Residential Loan Application handles joint intent directly. Section 1a asks whether the borrower is applying for individual or joint credit, gives the total number of borrowers, and has a line that reads "Each Borrower intends to apply for joint credit," with a space for each borrower's initials. That initials line is separate from the acknowledgments and agreements each borrower signs at the end of the form, so on paper it is the distinct method the comment asks for.

Digital applications are where it comes apart. In many point-of-sale systems, the initials box becomes a checkbox the primary borrower ticks for everyone, or a field that auto-populates when a second borrower is added, or a signature captured once and applied to every page. Each of those turns a co-applicant's own affirmation into something the first applicant, or the software, did on their behalf. The URLA still shows initials, and the record behind them shows nobody chose anything.

That gap gets wider when the application moves to a phone call or a chat.

## Where a Conversational Application Goes Wrong

When the [POS](/for-lenders/pos) agent takes an application, it is having a conversation. Borrowers say things like "my husband and I are buying a house" or "can my mom go on the loan." Each of those can mean several different things. The husband may be a co-applicant. He may be a non-borrowing spouse on title who will not sign the note. The mother may be a non-occupant co-borrower, or someone giving a gift, or someone the borrower thinks has to be on the loan for no reason the program requires.

A form forces the choice. A conversation lets an agent infer it, and inference is the failure mode. If the agent hears "my husband and I" and creates a two-borrower application, it has decided on the borrower's behalf that the husband is applying. If it asks the husband for his income documents a minute later, the lender now has his paystubs, his credit pull authorization and his name on the 1003, and nothing anywhere shows that he chose to apply.

So we built the joint intent capture as a rule in the [Rulebook](/how-it-works#rulebook) that the agent cannot route around, with four parts.

The agent never adds a second applicant on its own reading of the conversation. When it hears a second person mentioned, it asks a neutral question: whether that person is applying for the loan with the borrower, and it explains in plain terms what applying means, which is that both people's income, assets and credit are considered and both are responsible for repaying.

Each applicant affirms their own intent. The person who started the application cannot affirm for the co-applicant. The co-applicant gets their own invitation, authenticates as themselves, and affirms intent to apply jointly in a step that is separate from the screen where they later attest that the information is true. This is the distinct method the comment asks for, done by the person whose intent it is.

The affirmation is stored as an event in the [Loan Graph](/platform/loan-graph) with the applicant, the wording shown, the time, and the channel. It is written before any credit authorization for that person is requested, because the comment says intent has to be evidenced "at the time of application," and the credit pull is the point at which most lenders would say an application exists.

If a co-applicant is removed later, the agent records the removal as that applicant's decision, with the same detail, and recalculates what the file now needs.

## What the Agent Does Not Suggest

The spousal signature rule is the other half, and it is a question of what the agent says. An agent trying to help a borrower qualify might reach for the obvious lever: "If your spouse applies with you, you would have more income." Sometimes that is true and the borrower wants to know it. But a lender cannot require the spouse to join when the applicant qualifies alone, and an agent that steers borrowers toward adding a spouse in that situation is building the pattern the rule forbids, one helpful suggestion at a time.

So the agent works the other way around. It takes the application as the borrower presents it. If the borrower qualifies alone, it never raises adding anyone. If the borrower does not qualify alone, the decision about what to tell them goes to a licensed loan officer, which also keeps the agent on the right side of the [SAFE Act licensing line](/blog/safe-act-ai-mortgage-assistants-nmls-licensing-line). Under 1002.7(d)(5), when an applicant does not qualify alone, a creditor may require a cosigner or guarantor, but it may not require that the additional party be the applicant's spouse. The agent leaves that conversation to the loan officer and records that it happened.

Non-borrowing spouses raise a separate point. In some states a non-borrowing spouse has to sign the security instrument to make the lien enforceable, or to waive homestead rights. Regulation B allows that, at 1002.7(d)(4), because it is a signature to create a valid lien, not a signature on the note. The agent knows the difference because the Rulebook carries the state's requirement, so it can tell the borrower early that the spouse will sign the mortgage at closing without becoming liable on the loan. That one sentence prevents a lot of closing-table confusion, and a lot of "just add him to the note so it's simpler."

## Checking the Humans Too

Most joint intent violations are made by people, not systems. A loan officer on a call says "we'll just put your wife on, it's easier," and the application goes in with two borrowers. When we built call QA for Better Mortgage, the [Compliance Agent](/for-lenders/compliance-agent) went from four audited calls a day against a backlog of more than 1,000 to a full month of calls scored in under an hour, with hundreds of the lender's own SOP documents codified into the rules it scores against, and Regulation B was in that set alongside UDAAP, TILA, RESPA and Fair Housing. Language about adding a spouse is one of the patterns a full review catches that a sample of four calls a day never would. When a call shows a loan officer suggesting a co-applicant to a borrower who already qualified, the finding carries the timestamp in the recording and the rule, and it goes to the fair-lending officer.

The pre-close side matters too. Before closing, the [Pre-Close QC Agent](/for-lenders/pre-close-qc-agent) checks that every borrower on the note has a joint intent event dated on or before their credit authorization, and that every signature on the note belongs to a borrower. A note signer with no intent event is a finding, raised while the lender can still fix it, which is before the borrower signs.

## What Examiners Ask For

An examiner testing 1002.7(d) usually pulls a sample of joint applications and asks the lender to show, for each, where intent was evidenced and when. On a paper process that means finding the affirmation page in the imaged file. On a conversational process, the answer is an event log: this applicant, this wording, this time, this channel, before this credit pull. Sei AI stores that log on the loan in the [Evidence Trail](/how-it-works#evidence-trail), so producing it is a query rather than a search. Our post on [recordkeeping and retention for AI agents](/blog/ai-mortgage-agent-recordkeeping-retention-ecoa-trid-audit-trail) covers how long ECOA requires those records to be kept, and our post on [fair lending and AI agents](/blog/fair-lending-disparate-impact-ai-agents-ecoa-hmda) covers the wider testing.

None of this is new law. What is new is that the application is a conversation, and a conversation produces a joint application by accident far more easily than a form does. The fix is to make intent a thing each applicant does, on their own, at a moment the system can point to.

---

_Source: [https://www.seiright.com/blog/joint-intent-co-applicant-mortgage-point-of-sale-ai-regulation-b-1002-7-d](https://www.seiright.com/blog/joint-intent-co-applicant-mortgage-point-of-sale-ai-regulation-b-1002-7-d) · Sei AI_
