Robots Took Over My Mortgage Company: Why It’s So Hard to Get Help from a Human

But here’s the truth: a lot of these tools don’t work the way companies think they do. AI tools aren’t built like a traditional calculator. They are black boxes that hallucinate, meaning that they sound confident while being flat-out wrong and can’t explain themselves. And when they get something wrong in mortgages, real people pay the cost, not the robots and certain not the mortgage companies who deployed them.

What “black box” really means

AI models are often called “black boxes” because even the people who build or use them can’t fully explain how they arrive at a decision. These systems rely on massive amounts of training data and complex math to spot patterns, but the logic behind any single answer is usually hidden or impossible to trace. That makes it hard to know if an AI is making fair, legal, or accurate calls, especially in high-stakes areas like mortgages.

The CEO of a mortgage company explained how relying on AI to make high-stakes decisions about a home can be illegal if the AI cannot demonstrate how it arrived at any given decision. He wrote:

[Federal regulators require] any bank relying on AI. . . to justify how the system arrived at its conclusions, ensuring accountability and compliance. In high-stakes applications where life, property, or financial security are at risk, AI cannot operate on the principle that outcomes alone justify the methods — it must be explainable and auditable.

The black box nature of AI models can create dozens of problems. For example, studies have shown AI regularly discriminates against minorities. A study from April 2025 examined multiple generations of AI bots and found that they deny home loans to Black applicants more frequently than to White applicants who are otherwise identical. And when the AI did approve Black applicants, it charged them higher interest rates.

Hallucinations, too!

AI “hallucinations” are when the system gives a fluent, confident answer that’s simply wrong. This isn’t rare. As one 2024 study summarized in TechCrunch put it: “At present, even the best models can generate hallucination-free text only about 35% of the time.” Think about that. If your pharmists only gave you the right dosage 35% of the time, would that be good enough or would it be an attempted murder case in the making?

Worse, some researchers argue this problem may never fully go away. Formal results published in 2024 say it’s impossible to eliminate hallucinations in large language models, and recent coverage echoes that the issue stems from how these systems work. In plain English: the tech guesses the next words; sometimes those guesses sound right but aren’t.

Why this matters for mortgages: imagine you send a proper Notice of Error because your escrow shows the county taxes weren’t paid. A chatbot or auto-draft tool could “hallucinate” that the taxes were already disbursed and close your ticket, because it pulled or inferred the wrong thing from your file. Or it could invent a deadline you supposedly missed, misstate your eligibility for help, or auto-populate the wrong loan owner. In a high-stakes context like servicing, a confident wrong answer can snowball into fees, missed cure windows, or even foreclosure risk.

Where you’re most likely to meet the bots

• Phone trees and voice “assistants.” Lenders and servicers are testing voice bots that handle intake and routine questions before any human gets on the line. Even optimists acknowledges the risk of misinformation and the need to actively address it.
• Chat support. Servicers are adding “AI-powered” chat. A J.D. Power deep-dive found a strong consumer preference for human agents, and that transparency (if the consumer knows they are talking to a bot or a person) matters. When customers thought they were chatting with a human, they reported better outcomes.
• Document review and decision helpers. Inside the back office, AI already reads W-2s, calculates income, compares loan estimates, and drafts responses. Industry leaders say the point is to re-imagine workflows, not just bolt a faster script on the old process. This is another sign the shift to AI is deep and accelerating.

How these issues can hit a real homeowner

• Phone maze purgatory. You’re routed to a bot that can’t understand you, can’t verify your documents, and can’t escalate. Meanwhile, a payment problem or foreclosure timeline keeps moving. Regulators have warned that this kind of “bot wall” can block access to timely human help.
• Inconsistent instructions. You apply for a loan modification. The website bot says “upload pay stubs only.” The chat bot later says “you also need bank statements and a hardship letter.” A voice bot on the phone says “we’ll pull those automatically.” You follow one set of steps and get denied for “missing documents.” Conflicting AI instructions waste time, can sink a legitimate application, and make it hard to protect your home.
• Auto-drafted letters with auto-generated mistakes. If a company leans on AI to summarize your file or to draft a response to a Request for Informaiton or Notice of Error, it might confidently include the wrong payoff, dates, or owner information. and deadlines.

---

Your rights (and how to use them if a robot is in your way)

1. Ask for a human, clearly and early. Say, “I need a live representative.” If you’re in chat, ask “Are you a bot or a person?” If the system won’t connect you, note the date/time and keep screenshots. Always, wether it’s a human or a bot on the phone, take notes about what they say.
2. Use the magic words in writing: “Notice of Error” or “Request for Information.”
   Under federal law (Regulation X):
   • For a Notice of Error (NOE), the servicer must acknowledge within 5 business days and generally investigate and respond within 30 business days (with a possible 15-day extension if they notify you). Some specific errors (like payoff amounts) have shorter deadlines.
   • For a Request for Information (RFI), they must acknowledge within 5 business days and respond within set timelines (e.g., 10 business days for owner/assignee info; generally 30 business days otherwise, with a possible 15-day extension).
     
3. If you’re denied or your terms change, demand the real reasons. Lenders can’t hide behind AI. If they take adverse action, you’re entitled to specific factors, not vague boilerplate. Save the letter; if it’s generic, that’s a red flag.
4. Escalate when the bot blocks you.
   • File a complaint with the CFPB. Companies must respond through the portal; servicing issues are a major source of mortgage complaints each year.
   • Talk to a housing counselor or legal aid. If timelines are slipping toward foreclosure, get help fast.
5. Keep a paper trail.
   Save call logs, chat transcripts, screenshots, and copies of everything you send. If a bot gave you bad info that cost you time or money, documentation matters.

---

Bottom line

AI is already inside your mortgage company. Sometimes it’ll speed things up. Sometimes it’ll stall, guess, or gloss over your rights. If you hit robo-walls, ask for a person, use your written rights (NOE/RFI), demand specific reasons, and escalate. You shouldn’t need a computer science degree to keep a roof over your head.

---

About the Author

Angel E. Reyes is a former federal enforcement attorney at the Consumer Financial Protection Bureau and the Federal Trade Commission. After bringing enforcement actions against the largest U.S. companies, which resulted in over $100 million returned to consumers, he left the government to open Power to the People Law PLLC. This law firm focuses on protecting people’s homes and bank accounts.

Disclaimer

Informational only. Attorney advertising. Not legal advice.