The Worst Version of This Call
A woman calls the servicer about her late husband's mortgage. She is not on the loan. She never signed the note. She wants to keep the house, she is trying to find out what is owed and whether she can take over the payments, and the first thing the servicing script does is refuse to talk to her, because she is not an authorized party on the account. She calls back a week later and gets the same answer. In the meantime the loan goes further past due, a loss-mitigation window she did not know existed closes, and a house that could have been saved moves toward foreclosure because the person trying to save it could not get the servicer to acknowledge she existed.
This is the failure mode the successor-in-interest rules were written to stop, and it is a failure mode an AI servicing agent can reproduce at scale or prevent at scale depending on how it is built. An agent that treats "you are not on this loan" as the end of the conversation is an agent that turns a grieving family member away from the exact process meant to help them. We work with servicers on where the AI agent sits on these calls, and the design principle is the one the regulation embodies: the call from someone not on the loan is not a call to end, it is a call to route.
What Reg X Actually Requires
The 2016 amendments to the CFPB mortgage servicing rules extended a set of borrower protections to successors in interest, and the mechanics are specific. A successor in interest is defined at 12 CFR 1024.31 as a person to whom an ownership interest in the property securing a mortgage loan is transferred from a borrower, provided the transfer falls into one of the named categories: a transfer on the death of the borrower to a relative or joint tenant, a transfer to a spouse or child, a transfer resulting from a divorce or legal separation, or a transfer into an inter vivos trust in which the borrower remains a beneficiary, among others. These are the life events, death, divorce, inheritance, family transfer, that produce a new person with a stake in the property who was never a party to the loan.
The servicer's duties run through two provisions. Under 1024.38(b)(1)(vi), a servicer must maintain policies and procedures reasonably designed to promptly facilitate communication with any potential successor in interest, respond to inquiries, and identify the documents it needs to confirm that person's status. Under 1024.36(i), when a servicer receives a written request that indicates the person may be a successor in interest, it has to respond by providing the person with the information the servicer requires to confirm their status. The rule does not ask the servicer to take anyone's word for it. It asks the servicer to tell the person what confirmation requires and to process the documents when they arrive, rather than stonewalling at the front door.
Once the servicer confirms the successor, the status of the call changes entirely. A confirmed successor in interest is treated as a borrower for the purposes of the Reg X servicing protections, the loss-mitigation procedures, the error-resolution rights, the information requests, and under the corresponding Reg Z provisions receives the periodic statements and disclosures a borrower receives. Confirmation is the hinge: before it, the servicer's duty is to facilitate and confirm, and after it, the person the servicer would not talk to last month is a person with the full set of borrower protections.
Where the Agent Recognizes the Signal
The most important thing the agent does on these calls happens in the first thirty seconds, and it is recognition. A caller who says "my husband passed away," "we got divorced and I kept the house," "my mother left me the property," or "I inherited this house" has just given the successor-in-interest signal, and the agent has to hear it as that rather than as an authentication failure. The design decision we push for is explicit: those phrases are a hard interrupt. The agent stops running the standard identify-the-authorized-party flow, because the standard flow ends in a refusal, and starts the successor process instead.
Starting the successor process means the agent does three things. It acknowledges that the caller may be a successor in interest and that there is a defined path, which is the opposite of turning them away. It provides, or routes to a channel that provides, the list of documents the servicer requires to confirm status, the death certificate and evidence of the transfer, the divorce decree, the deed, whatever the servicer's confirmation policy calls for. And it opens a successor case in the system of record with the date of first contact, because the promptness the rule requires is measured from when the servicer first learned a potential successor was reaching out, and a call that ends without a case opened is a call the compliance record cannot show was handled.
There is a privacy line the agent has to hold while it does this. Before confirmation, the caller is not yet a confirmed successor, and the servicer's obligations to the account and to the original borrower's privacy still apply, so the agent walks the caller through the confirmation process and explains what is needed without disclosing account details it is not yet permitted to share. Walking them through the path is not the same as treating an unconfirmed caller as the borrower, and the agent has to do the first without doing the second.
The Line the Agent Absolutely Cannot Cross
Confirmation as a successor in interest is not assumption of the debt, and conflating the two is the mistake that creates real liability. A confirmed successor gains the servicing protections and the right to apply for loss mitigation, but they do not automatically become personally liable on the mortgage note, and the process by which someone assumes the loan and becomes obligated on it is a separate legal step with its own requirements. The Garn-St Germain Act protects many of these transfers from due-on-sale enforcement, which is why the successor can keep the property without the loan being called, but protection from acceleration is not the same as liability on the debt.
The agent has to keep these distinct in what it tells the caller. It can tell a confirmed successor that they may apply for loss mitigation and that the servicing protections now apply. It cannot tell them they are now responsible for the debt, cannot represent that they have assumed the loan, and cannot make the assumption determination, because assumption is a legal step the servicer's people own and getting it wrong in either direction, telling someone they owe a debt they do not, or telling someone they are safe when an assumption they think happened did not, is a representation the servicer is accountable for. The agent's job is to run the recognition and the confirmation intake and to route the assumption question to the humans who own it, the same discipline we hold across loss mitigation and the rest of the servicing desk: the agent runs the regulated intake, and the human owns the determinations that carry legal weight.
Why This Belongs on the AI Agent at All
A reasonable objection is that these calls are rare and emotionally difficult and maybe should never touch an AI agent. The rarity is exactly why they get mishandled. A servicing floor handles successor calls infrequently enough that many agents have never been trained on the confirmation process, so the default response to a caller who is not on the loan is the privacy script, which is the wrong response. An AI agent that has the successor recognition and the confirmation process built in handles the rare call the same correct way every time, which is more than can be said for a human queue where the process lives in a rarely-opened section of the manual.
What the agent changes is not the human judgment, the confirmation decision and the assumption determination stay with people, and the emotionally difficult conversation can and often should move to a person quickly. What it changes is the front door. The caller who says her husband died gets acknowledged and routed into the successor process on the first call, with a case opened and the required documents named, instead of being turned away three times while the loan slides toward foreclosure. That is the outcome the rule is trying to produce, and it is a recognition-and-routing problem, which is the kind of problem an agent is good at.
The Honest Read
A death, a divorce, or an inheritance produces a caller who is not on the loan and has a real stake in the property, and Reg X turns that call into a regulated process: recognize the potential successor, tell them what confirmation requires, process the documents, and once confirmed treat them as a borrower with the full servicing protections. The failure mode the rule was written to stop is the servicer that answers "you are not on this loan" and hangs up, and an AI servicing agent that runs the standard privacy script on these calls reproduces exactly that failure.
An agent earns its place by hearing the successor signal in the first thirty seconds, stopping the refusal flow, and starting the confirmation process with a case opened and the promptness clock running, while holding the privacy line before confirmation and the liability line at every point. At Sei, we build the servicing agent to recognize the successor call, facilitate the confirmation the rule requires, and route the assumption question to the people who own it, because confirmation is not assumption and no agent should ever tell someone they owe a debt they may not. The call from someone not on the loan is not a call to end. It is the call the rule cares most about getting right.
Pranay Shetty
CEO & Co-Founder