Can Generic AI Review a Mortgage Loan Purchase Agreement, and What Is It Missing?

Can a general-purpose AI model review a mortgage loan purchase agreement? No — not competently, and not because the model reasons badly. The obstacle is evidentiary. A general model learns from public text, and the provisions that decide a secondary-market trade’s legal economics are almost never written about in public. The model has not read the material, because the material is largely not there.

Why is this different from ordinary contract review?

Ask the open internet how a seller guide amendment right operates, or how an early payoff window is measured. Back come the aggregators’ own guides, vendor marketing, and agency job aids. Seller-side, provision-level analysis is close to absent.

A survey across the published insight libraries of eight firms active in residential mortgage finance — Mayer Brown, Alston & Bird, Hunton Andrews Kurth, Skadden, Cadwalader, Dentons, Ballard Spahr, and Weiner Brodsky Kider — returned practice-area descriptions, attorney biographies, and regulatory-update posts, and no analysis of how any of these provisions operates. These firms do this work daily. It is simply not published.

So the corpus behind a general model’s answer on an MLPA, an MSR purchase and sale agreement, a seller guide, or a warehouse facility is the counterparty’s paper and the vendor’s pitch.

What does a general model not know about a seller guide?

The 2003 correspondent purchase and sale agreement between Washington Mutual Bank, FA and Crescent Mortgage Services, Inc. runs five pages. Article I incorporates the seller guide “by reference in its entirety.” Article V leaves the seller’s repurchase obligation “as more fully set forth in the Seller Guide,” and Articles VI and VII do the same with suspension, termination, and duration. Section 8.13 settles the hierarchy: “In case of any inconsistency between this Agreement and the Seller Guide, the terms and provisions of the Seller Guide shall control.” Section 8.2 then provides that while the agreement “may not be amended except by an instrument in writing signed on behalf of each of the parties hereto,” the seller guide “may be amended or supplemented by Purchaser from time to time in Purchaser’s sole and absolute discretion,” and a seller who “registers, locks or delivers a Mortgage Loan after receiving notice of a proposed amendment . . . shall be deemed to have agreed to such amendment.”

Five pages deferring nearly every operative term to a document the model has not been given, subordinate themselves to it, and let the counterparty rewrite it unilaterally. Nothing in the output will read as incomplete. What an aggregator can change without consent takes that right apart.

Which provisions does it read in isolation and misstate?

Start with the early payoff window. Pennymac’s delegated seller guide provides that an early payoff “exists when a Mortgage Loan is paid in full or when a curtailment in excess of 30% of the original principal balance occurs within 180 days after the Funding Date.” The same guide separately reaches agency-eligible loans “paid in full within 120 days after the Funding Date,” cross-referencing Fannie Mae’s Selling Guide. That Fannie provision measures from somewhere else: “any loan that pays off within 120 days from the whole loan purchase date or the MBS issue date, as applicable.” Same day count, different starting event. Neither document announces the divergence; no market standard supplies it. EPO premium recapture sets the published windows side by side.

Then the knowledge qualifier. Fannie Mae “may also require repurchase or a make whole payment if any warranty the selling lender made is untrue and, if the remedies framework applies, qualifies as a significant defect, whether or not the lender had actual knowledge of the untruth.” The next sentence supplies the exception: no such demand “will be made if the warranty specifically states that a violation does not exist unless the lender had actual knowledge of the untruth and the lender has no such knowledge.” Quote either alone and the exposure is misstated. Materiality and knowledge qualifiers, survival periods, sole-remedy clauses, and indemnification caps operate on one another; summarizing any one in isolation produces a confident wrong answer.

What a counterparty will concede appears in no document at all. That is market knowledge.

What has already gone wrong when lawyers relied on it?

In *Mata v. Avianca, Inc.*, plaintiff’s counsel filed a brief citing decisions that did not exist. The court found that counsel “abandoned their responsibilities when they submitted non-existent judicial opinions with fake quotes and citations created by the artificial intelligence tool ChatGPT, then continued to stand by the fake opinions after judicial orders called their existence into question,” and imposed a $5,000 penalty jointly and severally.

The detail that carries outside litigation is in the findings of fact. Counsel asked the model whether the cases were real. “ChatGPT responded that it had supplied ‘real’ authorities that could be found through Westlaw, LexisNexis and the Federal Reporter.” Asked to check itself, it confirmed itself.

The court was careful about the general point: “Technological advances are commonplace and there is nothing inherently improper about using a reliable artificial intelligence tool for assistance. But existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings.”

*Mata* was a court filing, where a fabricated citation is eventually caught by an adversary or a judge. A misread survival period has no adversary reading it. It surfaces years later, in a demand.

Where is AI genuinely useful here?

In a great many places — a first cut of an operational policy, a thousand-page agency guide compressed for a credit committee, a diligence index organized, administrative backlog cleared. A general model handles all of that well.

The line is not the technology. It is whether the output defines a trade’s legal economics. Rep and warranty drafting, sole-remedy clauses, survival periods, indemnification caps, and MSR purchase and sale structuring sit on the far side of it. Where AI belongs there, it is AI trained on the deal documents and supervised by counsel who negotiated them — the premise behind DealGPS.

What should you actually do?

Practitioner takeaway. Before a model’s read of a purchase agreement informs a term sheet, put one question to it: what document does this agreement incorporate that you have not been given? If the answer is the seller guide — and it usually is — the review has not started.