Mortgage Regulation Outlook 2025 – LIVE BLOG from MBA Policy Conference

Kicking off the MBA Mortgage Regulation 2025 Live Blog with some notes on policy players in Washington.

Last week the nominee to lead CFPB Jonathan McKernan was reassigned by Scott Bessent to a big role at Treasury.

This implies Russell Vought from OMB who’s been leading the dismantling of CFPB is likely to continue in this role.

New FHFA leader Bill Pulte is expected to do a lot to modernize Fannie Mae and Freddie Mac.

The MBA is also working with FHFA to share ideas about how an eventual removal of Fannie Mae and Freddie Mac from conservatorship may play out.

The main goal is fair pricing and easy access to mortgages for consumers.

Pulte will join MBA next week in NYC for MBA secondary.

Trigger lead bill limits trigger leads to institutions who already have relationships with consumers.

Isaac started this panel with some key notes:

– Treasuries are the cleanest dirty shirt in global bonds so even with all the policy volatility, America and rates overall look decent.

– We should be prepared to die a thousand deaths as each policy plays out. It’s just the style of this political era.

Ethan on tax bill:

– No tax on tips and overtime, and benefits for seniors are Trump Administration priorities going into the bill but would expire at the end of his term.

Pete on mortgage and housing tax bill items:

– Mortgage interest deductions stay.

– They’ll add more opportunity zones.

On February 18, Trump issued executive order to reign in independent agencies.

This means “CFPB and FHFA are independent in name only.”

As for the Fed, they’re likely maintain monetary policy independence, but may become less independent on regulatory and supervisory activities.

Isaac said Mark Calabria at OMB looks after financial regulatory infrastructure.

As such “Calabria may be one of the most important people in financial regulation.”

Also Jonathan McKernan at Treasury now will be extremely influential in financial regulation. And as Pete noted, McKernan will also work extremely closely with Bill Pulte at FHFA.

What effect will Trump policies have on home builder costs.

Ethan noted a stat from NAHB who estimated that tariffs may increase home building costs by about $10,000 per home.

This is not market breaking.

But the chaos and volatility of the tariff policies are what make this hard to predict.

Tariffs were 2.5% on average and now they’re 10% on average.

Just based on this, it’s 4X and we need to watch closely to see how this hits inflation numbers, which will keep rates higher for consumers.

And it may also impact current builder predictions of building costs.

Will Fed hold, hike, or cut in 2025?

Nobody knows.

With tariffs, 10% is the new zero percent.

Another way Bessent has suggested keeping rates under control is the supplementary leverage ratio: let banks buy treasuries which will lower 10yr Note by 30-70 basis points.

KEY/FUNNY: The UK accord keeps tariffs low on gin and scotch. 🥃

Liklihood of GSE release from conservatorship:

Treasury will be the leader of this effort.

McKernan will lead this but will be careful and needle won’t move until 2026.

How does sovereign wealth fund relate to ending conservatorship.

There’s $330 billion on top of capital stack that’s politically fraught. The government owns 80% of common stock.

You could maybe fund sovereign wealth fund with government stake in GSEs.

Does that mean GSEs are still government backed? Sort of, but less explicitly.

KEY: Moving shares from government to sovereign wealth fund doesn’t release GSEs from conservatorship.

Fannie and Freddie both still have a long way to go to hit capital levels to qualify for release from conservatorship.

So far, the administration has signaled that lowering capital requirement levels isn’t on the table.

If McKernan had stayed at CFPB, he may have kept a functional structure.

With Russ Vought, they’re likely to trim agency down to 200 people.

Then there might be room for RESPA reform and loan officer compensation reform — hopefully — but they wouldn’t have the resources for much more.

Panelists posited that Calabria is also quietly running the CFPB.

And also that Calabria is maybe the de facto U.S. housing czar MBA has talked about, even if it’s not a named role.

Our Mortgage Regulation 2025 Live Blog continues with former regulators turned professors and deep policy wonks. This will be a (civilized) debate between two political perspectives.

Christopher kicked off panel noted CFPB is NOT the independent agency critics claim it to be.

The director can be removed. And it doesn’t have any more inherent independence than other agencies.

BUT… funding structure is part of the criticism the CFPB gets. More on that later.

Todd was involved with CFPB in Trump’s first administration under former director Kraninger.

Todd talked about how bureaucratic the FTC was when he worked there. The main issue was coordination among agencies.

Todd said his view was there was no need to create the CFPB and it should’ve all gone to FTC.

But here we are. And Todd said he agrees with Alito that there are almost no checks on CFPB power.

Todd said he knows former CFPB director Chopra and considers him a friend but he was granted the ring of power and overreached with it.

Todd said Congress needs more say with CFPB.

Christopher “strenuously disagrees” with Todd on governance.

He said the judgment of Congress is involved.

The crux of the debate here is that CFPB is funded by Fed.

Todd (and those aligned with him) believes this renders Congress powerless.

Christopher (and those aligned with him) disagrees because he said Congress can direct Fed how much to fund.

Todd said the Congress appropriations process should be like Christopher decides, but instead the President can overrule that.

Now the CFPB director is removable (by the President) which is different from how it began.

Todd said bipartisan agencies were historically more stable, but now they’re just 3-2 votes on anything that matters.

He said many of the board chiefs of those bipartisan agencies are former Democrat or Republican who’ve become really partisan.

Christopher noted Trump has fired 2 FTC commissioners before their terms are done — to make the point that it’s all partisan, and the quest for bipartisanship remains important.

He also jabbed at Todd on his Federalist Society “crystal ball.”

Christopher said “maybe the Supreme Court is waking up to corruption.”

Todd agrees with Christopher in the “sea change” in Trump’s views on the administrative state.

As first they wanted independent agencies. Then they decided the “deep state” — including independent agencies — undermined the Trump agenda, and need to be brought into presidential control.

Both Todd and Christopher talked about their dissatisfaction with TILA-RESPA integrated disclosures (TRID).

This is a reference to TRID consumer disclosures for mortgages that went into effect October 2015 and dictate what forms lenders must use to quote mortgage rates and fees to consumers.

It’s actually pretty well accepted and streamlined but these two policy wonks say they’re not great.

But they also both said it would be tough to tackle this again.

For those of us in the trenches — disclosing loans to consumers — before TRID, I disagree with both Christopher and Todd.

I think the TRID disclosures are quite transparent and also provide ample time for borrowers to make informed decisions (the TRID process has timelines and fee adherence rules associated with first Loan Estimate disclosure and final Closing Disclosure).

Before the Loan Estimate and Closing Disclosure TRID process, there was the Good Faith Estimate up front and the HUD-1 at closing.

These 2 disclosures were totally different in format and confusing to consumers. This led to great inefficiency in the lending community because you’d have to re-explain everything to borrowers at the end of the deal to make sure everything was aligned to the original quote.

===

Christopher commended the CFPB complaint database and enforcement process.

He said trying to dismantle the agency will leave us in chaos.

He said the president can’t just rewrite law that created CFPB (however the administration is in fact dismantling the agency).

Todd on success metrics for the CFPB.

He commended Christopher for helping to get the agency up and running under director Cordray.

He repeated that director Chopra “went hog wild” with regulation through enforcement.

This was also a major mortgage industry criticism of the Cordray-era CFPB.

Todd gave credit to Chopra for pushing innovation and competition.

He said there’s still no core structure for fintech to promote innovation and competition.

This is a good question for the future of a drastically reduced CFPB:

Can the CFPB play the role of supporting fintech innovation?

Christoper doesn’t like the label of “regulation through enforcement” and says it’s just: enforcement.

I disagree, especially as it pertains to the Cordray era. Here’s a description of how regulation through enforcement became a monicker in that era.

And here’s an excerpt of that history:

The CFPB was created by Dodd Frank in July 2011, and its first director Richard Cordray ran it from 2012 to 2017, during which time he’d issue enforcements known as consent orders.

Each consent order referred to laws but didn’t necessarily say lenders had broken these laws. Lenders just received fines that would hobble but not necessarily break them. And of course they’d be subject to specific corrective actions on issues like loan officer compensation, affiliated business agreements, etc.

Cordray, in public speeches, would say all industry players were expected to abide by corrective actions in consent orders or be subject to the same enforcement fate.

This came to be known as “regulation through enforcement” and like the crypto industry today, the mortgage industry was super pissed.

But the mortgage industry took its lumps and is quite stable now.

Law clarity now outweighs enforcement actions. But it took an entire era.

===

Todd correctly pointed out that the CFPB was unfairly blamed for debanking crypto.

Actually Chopra was supportive of blockchain innovation.

Other agencies were much more restrictive to block gain and crypto.

On the topic of supervision, Todd said this needs to be viewed as principles vs. rules based philosophy.

This is a layered remark worth contemplating.

But the Mortgage Regulation 2025 Live Blog goes onto the next topic for now…

Judge Joshua Wilson is a U.S. District Court judge in the Eastern District of Philadelphia. This is an open panel on today’s litigation trends, and hopefully that’ll include some notes on AI.

===

JUDGE WOLSON BIO

Judge Wolson was nominated to the United States District Court for the Eastern District of Pennsylvania in May 2018. He was renominated in May 2019, was confirmed by the Senate on May 2, 2019, and received his commission on May 28, 2019. Judge Wolson earned his J.D., cum laude, from Harvard Law School, B.A., magna cum laude, from the University of Pennsylvania. After graduating from law school, he served as a law clerk for Hon. Jan E. DuBois of the U.S. District Court for the Eastern District of Pennsylvania. He then maintained a commercial litigation practice, including antitrust, RICO, intellectual property, class action, First Amendment, and commercial contract disputes, first as an associate with Covington and Burling in Washington, D.C., and then as a partner with Dilworth Paxson in Philadelphia.

===

Post COVID he said juries are “more generous” than before.

On AI, he said they use it in test mode.

You can build workflows for — as an example — how to enter a guilty plea.

He can take the plea memos and agreements and ask AI to summarize so he can go faster.

He likes AI for preparing jury instructions.

On the lawyer side, he cautioned against using AI like Google.

But you can use it to analyze and summarize big packs of docs and data.

If you’re looking at a case that involves tech and circuits, you can ask AI to explain some of these case technicals to you.

===

Should media mention what president a judge is appointed by?

Wolson said no.

He said judges should be judged by their decisions.

Judges are not what a presidential appointment “would theoretically imply.”

Next up in the Mortgage Regulation 2025 Live Blog is marketing topics.

RESPA is about making sure lenders and other real estate service providers aren’t creating higher costs for consumers by paying each other for referrals.

A “thing of value” is a key legal definition for this.

So is a Like on social media a thing of value?

Maybe, maybe not. But if it’s paid, it likely is.

Hard to determine, but this is what regulators — and lawyers trying to protect lenders and other service providers — are looking out for.

Another example is loan officers saying on social how they help Realtors promote their listings — or actually promote their listings.

This is considered a thing of value by regulators and lender compliance teams.

Here are some good slides on these topics:

The FTC rules on fake reviews are even more important in the AI era.

As you can see on the slide below, it’s on the company to take the review down.

But it’s a tough balance because there are also rules to prevent suppression of negative real reviews.

===

2024 Case Study: Social Media Kickbacks

Allied Title & Escrow

Yacht parties for real estate agents hosted by title company were shown on social media.

This was used to allege a thing of value was given to the real estate agents.

The agents did pay for this event, but it did result in a RESPA violation and a fine.

===

2017-2025 Case Study: Fair Lending Violation Through Podcast Content

CFPB alleged Chicago mortgage lender Townstone Financial had podcast content that was discriminatory.

The CFPB said this content created public perception damage about certain neighborhoods by talking about crime stats and using slang that CFPB interpreted as incendiary.

===

There are also lots of case studies about loan officers getting hit for RESPA violations after third parties posted positive comments about the loan officer that imply the loan officer offered a thing of value.

The use of AI on social media makes RESPA compliance patrolling tougher.

Automatically generated AI audio and video can inadvertently bring in protected music or images.

Ideally, and paradoxically, you can also use AI or some other tech to ensure compliance.

===

Kris Janssen at Rocket on:

Back To Basics: Leveraging Compliance Management Systems (CMS) In Marketing

Select key components for lenders include:

-Establish and communicate marketing compliance obligations for all team members.

-Actively manage vendors you use.

-Have a consumer compliant response process.

-What is your corrective action for violators?

-Clearly defined repositories to review pre-approved marketing materials.

-Marketing compliance training tailored to different team members based on how they use the brand publicly.

===

Here’s a good rundown of key laws and regulations applicable to marketing.

===

Federal laws require you to retain marketing materials for 2 years.

But states can require longer periods.

Examples: Minnesota requires 60 month (5 year) record retention and Virginia is 3 years.

Each state will also have specifics on exactly what and how much you need to retain.

Patti Dietz at Lennar Mortgage said they have a repository of pre-approved content for loan officers.

She cautioned on these forms of marketing:

The quest for great email subject headers and website banners often leads to regulatory risk.

Teaser ads as popups on the web can also incorrectly show rates with no APRs.

The internet — including social media — is forever.

It’s faster than ever to post and then everything else stays out there forever.

This makes the job of compliance officers harder because marketers and salespeople are often just posting off the cuff.

She gave a good example of a loan officer making an innocuous post like “happy holidays everyone” and a dissatisfied customer chimes in on comments with a complaint about their loan. These cases need compliant rapid response strategy.

She said for frequent posters, it’s good to:

– have either private social accounts

– recommend disclaimers for personal posts

– focus on sharing approved company content and/or only sharing content from official sources

This session ended with a quick rundown of state enforcements on marketing.

Prevailing state enforcement themes are about lazy, undisciplined language in marketing like:

– Best rates

– Best loans for veterans

– “Equal opportunity lender” disclaimer used by a broker who’s not a lender

That last one seems nitpicky but it sums up the entire theme of this session.

You must be disciplined about marketing language and the kind of content you post — which is much harder in the social era.

But even with a lighter enforcement tone currently in Washington, the states are coming at this topic as strong as ever.

The Mortgage Regulation 2025 Live Blog will now cover loan officer compensation, one of the hottest regulatory topics since the creation of Dodd Frank and the CFPB.

Even with the CFPB being scaled down, the loan officer compensation (LO comp) rules hold.

The basic premise of LO comp is per-loan compensation isn’t prohibited. It must be some percentage of total loan production.

These rules went into effect after the financial crisis and Dodd Frank created the CFPB which enforces it.

Before then, LOs could determine their comp on a per loan basis.

This was deemed unfair to consumers, and today’s rules compensate loan officers more like financial advisors get compensated — as a percentage of all their business, not single transactions.

One of the big questions now is whether the states will fill the CFPB void.

If they do, they may add former CFPB staff members.

Opportunistic plaintiffs’ lawyers may also start class action plans.

Case Study: Compensation Based On Lead Source

Are different compensation levels allowable for different types of lead sources?

For example, can an LO get paid more for self-generated leads and less for an internet lead given to them by their company?

This is permissible as long as prices of different lead sources are controlled centrally by the company and the LO has no ability to price an individual loan in such a way that changes their comp for that specific loan.

This pricing methodology per lead type must be applied up front because you know the lead source up front. If the lead type changed late in the loan process that’s a red flag for regulators.

===

Case Study: Compensation Based On Loan Type

Are different comp levels allowed for different loan types?

For example: can an LO get paid more for Non-QM loans or business loans, and less for conventional loans?

In short, it’s permissible but it runs the risk of a loan officer potentially suggesting higher comp loans to borrowers.

This steering activity is illegal.

But most loan types are so much different — because they exist for different borrower needs — LOs are less likely to do this.

But if they do, their company would likely take immediate action. And it creates serious licensing and career risk for LOs.

===

Case Study: Comp Based On Product/Intake Channel

Lender could view different loans requiring different levels of work and expertise.

Examples: purchase vs. refi loans, or specialty construction loans.

This is also permissible for lenders to set up different comp levels for these.

The rule also says you can pay differently for existing customer (refi) vs. new customer (purchase).

But this case study is ripe for enforcement challenges — especially from states.

Lenders will get uneven guidance and uneven enforcement from state regulators on this one.

Case Study: Reassignment Of Loans Based on Comp

Example: a producing branch manager assigns himself higher paid loans and his team lower paid loans.

This is a huge risk issue.

And a terrible manager.

===

Case Study: Paying Producing Managers

Example: flat rate mortgage pays producing branch manager on their own production plus override on branch, minus expenses of the branch.

Or flat rate mortgage pays LO on their production plus team override, minus expenses of the team.

These are both permissible, but they must be extremely codified by the company.

The smaller the teams or production in these cases, the more risky this model this is — because the expenses are then really closely tied to loans.

===

Case Study: Branch Manager Comp Changes By Product

Example: Branch manager gets higher or lower basis point overrides on non-QM or conventional loans.

Not permissible.

===

Case Study: Comp By Geography

Example: lender determines different risk factors for properties in different states.

If this is price sheets based on states, it’s permissible. If it’s price sheets LOs licensed in those states can change, then it’s not permissible.

===

Case Study: Comp For Dual Role as LO and Realtor

Example: a Realtor is also an LO and when selected as agent, she also does the loan, and gets a commission on both.

It’s doable but very fuzzy, and very high risk. It’s doable but very risky, and at the state regulatory level, there’s even more risk.

Another layer: what if she credits $1000 of real estate commission to buyer if they do loan with her.

Same fuzzy/risky notes above apply.

===

Case Study: Originated vs. Brokered Loan

Example: lender pays higher basis points for banked loans in house and lower basis points for brokered loans.

There’s a strong argument for this to be permitted to allow LO to place consumer in best possible loan.

But this is risky for the lender because their comp model is incentivizing LO to keep loans in house.

===

Case Study: Brokered Loans

Example: broker takes loans to 5 lenders who all pay differently.

This is permissible because broker doesn’t control what their different lenders pay them.

As for whether the broker decides to pay its LOs differently based on what each lender pays the broker manager, this is more risky.

Next up on the Mortgage Regulation 2025 Live Blog is another critical policy issue for lenders: RESPA.

RESPA is the real estate settlement procedures act which is designed to prevent different housing services providers from incentivizing each other for business referrals.

It’s not only enforced for written or oral agreements, it can also be enforced based on behaviors of service providers.

There has to be a residential mortgage loan to fall under RESPA Section 8.

Here’s a good overview of this.

Penalties for RESPA violations are steep.

It’s a criminal statute.

Criminal: $10k fine and one year prison term.

Civil: 3x amount of any payment for referrals.

Here are some recent Federal and state RESPA actions:

There’s a good CFPB FAQ on RESPA that explains how they define “thing of value.”

Of course all of this is changing with the new administration.

But the states will remain vigilant on RESPA issues.

Also removing guidance from CFPB — and the safe harbor it provides — actually creates more risk for lenders.

There’s certainly some near-term enforcement relief from a scaled down CFPB, but RESPA is still law.

Case Study: Mutual Referral Scenarios

Real estate brokerage A refers clients to brokerage B, and B pays referral to A.

This is allowed and there are forms in most local MLSs that brokerages use to disclose and codify this.

A second part of this scenario is Brokerage B also refers clients to title companies and lenders affiliated with Brokerage A.

This is allowed with the right disclosures.

===

Case Study: Marketing Services Agreements

It’s allowable for lenders and real estate companies to have marketing services agreements.

They have to be really specific.

Here are details on permissibility.

So lender can contract with a real estate brokerage for digital marketing.

They can also do it for physical marketing being handed directly to consumers as long as there’s an audit trail of all materials and fees. Though lawyers don’t love this — it’s defensible but they don’t prefer lenders do this.

The question is how much is the fee — how do you value it?

Ideally it’s a third party valuation but it doesn’t have to be.

Lawyers worry about under payment when referrals are going two ways. But if referrals are going one way, it’s less of a risk to pay less.

===

Case Study: Digital Mortgage Comparison Shopping Platforms

Steering is the issue here.

See briefing below.

Affiliated Business Arrangement overview below.

This is about jointly operated businesses between two entities, like a lender and a real estate brokerage — aka joint ventures.

These are allowed, but we don’t have guidance on sufficient capitalization.

Attorneys on this topic come up with guidance on how to capitalize these joint ventures.

Affiliated businesses aka JVs are also a target of state regulators so caution is warranted on all governance.

I’m kicking off day 2 with this panel.

Will come back later and do a summary on it.

Below are a handful of slides that summarize state of insurance market.

Insurance market as a whole is profitable but homeowner insurance market is dragging most of the sector down.

It’s because of more clams and premiums too low.

This is why premiums have been rising so much.

This slide below is notable.

It says we don’t have a long-term insurability crisis, but rates being too low to pay increased claims.

Insurance industry capacity should improve as higher rates rebuild ability to pay claims.

But the crux is of course that it creates home affordability issues.

===

Here’s how the insurance issue plays out on the ground for mortgage borrowers and lenders.

If a storm created an insurance claim to replace a roof that costs $15k and insurance will only pay half, borrowers often tell mortgage servicers they can’t pay their mortgage for X months because they need to replace their roof.

So it becomes a lender problem to solve.

===

A positive note the panel pointed out:

Resilience is growing.

Better rebuilds after natural disasters are surviving subsequent storms.

This leads to better mortgage performance because of fewer repair and replacement claims.

AI in financial services has a few key emphasis areas:

Fraud detection.

Credit scoring.

Customer service.

Cyber security.

Details summarized below, along with benefits and challenges.

AI risks are most often from models trained on incomplete and insufficient data.

Regulators want:

Fairness and accountability.

Things to be clear for consumers.

Protection of consumer data.

QUOTABLE: talking to regulators about AI is like talking to your teen about browser history — it’s uncomfortable but necessary.

Regulators don’t know enough about AI yet, and it’s up to mortgage industry participants to educate them.

Lenders need to identify technical-but-articulate members of their tech teams to talk to regulators.

===

KEY AI FEDERAL REGULATOR LIST

See below for a list of all Federal regulators, and some actions.

The outlook for AI regulation in 2025-2026 is summarized below.

The main risk is that if Federal regs don’t get organized quickly, there are many state regs that will come in and conflict but still be required for lenders and servicers.

Congress wants states to back off, but that’s a tall order when you see the spike in state AI legislation in the chart below.

And here are the states where AI legislation is most active through 2024.

And here’s a map of how much state AI legislation has proliferated in 2025.

As of early 2025, U.S. state legislators have introduced 781 AI bills.

This is more than all of 2024.

The slide below shows what topics are hottest in these bills.

Main themes in state AI bills proposed include:

Consumer protection.

Bias and discrimination.

Deep fake and synthetic content regulation.

Government AI use.

Limiting AI in youth education.

===

Main categories of state AI regulation.

Broad AI law that applies to all industries.

Really narrow focused AI.

In Colorado for example, the Anti-Discrimination in AI Act is broad.

It focuses on education, employment, financial services and lending, healthcare, housing, insurance, and legal services.

The slide below shows limited exemptions for financial services.

This Colorado bill doesn’t go into effect until February 2026.

This was to maybe get refinements to the bill before then, but panel said it doesn’t seem like it’ll change much.

The Colorado Anti-Discrimination In AI Act has duties specific to developers and deployers.

These are summarized below.

California has elected 2 AI bills.

First:

SB 942 requires watermarks on AI generated content as of January 2026.

Companies must include AI detection tool, so anyone can know who created the content.

If this isn’t provided within 96 hours then the license can be revoked.

Second:

AB 2013 says developers must provide details backgrounds and source lists used for their datasets.

California Privacy Protection proposed regulations include regulating automated decision making technology.

This has significant potential impact for lenders.

They added a sub-definition for lending decisions.

The comment period for this one closes June 2, 2025.

This is an example of where new AI regs can undue AI for automated underwriting mortgage approvals that’s been in place for decades and required by government backed Fannie Mae and Freddie Mac for lenders to use.

The MBA understands this and is advocating to help all state and federal regulations consider that if government backed underwriting processes are required for lenders, then new laws shouldn’t overwrite this.

===

As for Utah AI laws, they’ve taken a slightly lighter approach.

These are summarized below.

Data protection legislation has been picking up steam.

There are now 20 state privacy statutes enacted to date.

18 states have the right for consumers to opt out of automatic decision making.

Many of these have GLBA exemptions (here’s what that means)

As for Federal data privacy regs, there’s no real updates to laws for AI, but more guidance.

There’s a proposal for moratorium on enforcement of state laws that restrict development of AI.

It’ll move to House floor soon. But 40 state attorneys general have just sent a letter to Congress opposing this moratorium. Their case is: let the states regulate as they see fit.

As for pre-AI era regs, those are all in place, including:

Gramm Leach Bliley Act – GLBA

Fair Credit Reporting Act – FCRA

Telephone Consumer Protection Act – TCPA

Unfair, Deceptive or Abusive Acts or Practices – UDAAP

===

Key pragmatism note:

These last 2 panels are largely about all the potential risks of AI from a regulatory standpoint.

But the main safeguard is that if humans are involved in communicating with consumers and doing so following all of today’s financial services and lending laws at the Federal and state levels — which is required practice for mortgage lenders — then it’s generally business as usual.

AI can help your process in a compliant way if the above paragraph about your practices is true.

===

Another note on Colorado AI Act (which is covered in previous panel notes above):

Colorado AI Act is so broad, they had to specify that a calculator isn’t AI.

This is why lenders are so concerned about it. That and the fact that multi-state plus Federal AI policies make compliance difficult.

Will this make cost-per-loan reductions more difficult because compliance expense goes up?

Something to watch as this plays out.

===

Training data and consent risks:

Risk: LLM training requires huge datasets often from
public or semi-public sources.

This is something you must have controls in place so you’re not violating privacy laws.

Risk: Will AI make the wrong decisions?

This is mitigated by having humans in the loop.

Emerging Risks:

Unsupervised agents. As agentic AI proliferates, will it take over processes without controls built in?

MCP, A2A, and other integration points. Emerging integration standards still lack proper authorization mechanisms and broadly widen how models can improperly ingest and use data.

What is AI governance?

Here’s the definition:

A system of rules, practices, processes and technological tools that are employed to ensure an organization’s use, development and commercialization of Al technologies aligns with the organization’s strategies, objectives, and values; fulfils legal requirements, and; meets principles of ethical Al followed by the organization.

===

As for emerging standards on governance, here’s a good visual:

And here’s a link to read more about the NIST AI risk management framework.

Super helpful info on this site for NIST, which is National Institute of Standards & Technology, a non-regulatory division of the Department of Commerce.

===

3 AI specific privacy issues:

1. Converting anonymized data to identifiable. Is the model violating privacy by doing this?

2. Reduced data input awareness. Where did the model get the data?

3. A lack of AI privacy policy maturity. Are policies keeping up with the pace of innovation?

Every lending org will be different on their risk tolerance on these issues.

On items 1 and 2, it’s strongly recommended to classify, identify, and inventory all enterprise data.

And of course there’s a meta answer to this:

You can leverage AI to solve these AI data integrity and privacy issues.

That’s about as good of a place to end this AI post as I can think of 😂.