The Documentation Is the Risk

How AI Is Rewriting Fraud in Commercial Real Estate Lending

Imagine this conversation with AI: “Claude, create a guarantor personal financial statement as of 12/31/25 for Guarantor John and Jane Doe with a Wells Fargo bank account of $34,762.27, a 4-BR home in Teaneck, NJ with a value of $2,995,000 with a mortgage held by Citibank, NA of $948,459, 4800 shares of Alphabet, Inc., a 2020 Jeep Wrangler with a value of $28,000 and…..”

Thanks to AI, it is now possible to rapidly fabricate a high-quality, lender-friendly loan package capable of deceiving even experienced underwriters. A complete set of trailing 12-month operating statements, rent rolls, tenant leases, borrower financials and entity documentation – each with figures that reconcile across every page – can be generated using widely available AI tools by anyone with a laptop and working knowledge of commercial real estate (“CRE”) underwriting. These materials are professionally formatted, internally consistent, and, in many cases, virtually indistinguishable from legitimate reporting through manual review alone.

However, the same technological advances enabling this risk are also transforming how it is managed. AI and advanced analytics can surface inconsistencies, detect anomalous financial patterns, analyze document structure and language characteristics (including writing indications such as burstiness and perplexity),¹ and identify signals that may indicate manipulation or synthetic content. Together, these capabilities provide a more scalable and robust approach to fraud detection than traditional review methods alone.

Commercial real estate investors and underwriters can no longer rely solely on traditional diligence and surveillance methods to manage risk. As the fraud landscape evolves, these approaches must be augmented with advanced analytics and AI-driven tools that enable more comprehensive, scalable review. When integrated with established underwriting practices, these capabilities enhance the ability to identify inconsistencies, detect emerging risks, and respond more effectively to increasingly sophisticated forms of fraud.

Old Fraud, New Economics

The technology to implement fraud is already in the hands of bad actors. That capability alone elevates the risk for every institution with exposure to commercial real estate credit. This is true not because AI has invented new fraud schemes – the underlying methods of overstating income, concealing liabilities and inflating valuations have always been a risk for lenders – but because it has collapsed the cost and time of executing them convincingly. S&P Global estimates that the CRE maturity wall will peak at $1.26 trillion in 2027, up from $950 billion in 2024² – and for credit investors navigating that wave, the integrity of borrower-reported data can no longer be taken at face value. The traditional tools used for verification are no longer sufficient.

Case Example: AI-Generated Documentation and Large-Scale Fraud Risk

In 2026, Commonwealth Bank of Australia reported approximately A$1 billion in suspected fraudulent home loans to regulators following a review of lending documentation.³

Public reporting indicated that certain applications included potentially manipulated or synthetic documents, with growing concern around the use of generative AI to produce convincing financial records and borrower information.

The suspected activity involved doctored applications and falsified supporting materials submitted through lending channels, in some cases leveraging stolen identities and digitally generated content.⁴ The issue was not the novelty of the fraud itself, but the quality and internal consistency of the documentation – reducing the discrepancies lenders have historically relied on to identify misrepresentation.

The case underscores a broader shift: documentation that appears complete, consistent, and professionally prepared can no longer be assumed to be reliable without independent validation.

Fraud in CRE lending, like with many private companies, has often centered on the same basic problem: borrowers control nearly all of the financial information that lenders use to make credit decisions. Rent rolls are self-reported. Operating statements are borrower-prepared. Documents are transmitted by prospective borrowers or their agents. That information asymmetry has existed for as long as the asset class has, and sophisticated lenders have developed frameworks to manage it – site inspections, third-party appraisals, review of Argus models, and reliance upon experienced underwriters who develop an intuition when the numbers “feel” or “smell” wrong. Those safeguards are antiquated and insufficient.

What those frameworks relied on, often without explicit acknowledgement, was friction. Fabricating a convincing financial package used to require time, expertise and attention to detail. Documents produced by different parties at different times inevitably contained small inconsistencies – a rent roll that didn’t quite match the operating statement, logos that looked manufactured or photocopied, an expense ratio slightly outside market norms, or a lease abstraction with formatting artifacts. Those inconsistencies were the threads that, when pulled, unraveled fraudulent submissions.

AI eliminates that friction. Generative models can produce entire document sets from a single set of assumptions, ensuring that every figure, every ratio, and every cross-reference ties perfectly. The internal consistency that once served as a hallmark of legitimate reporting has become trivially reproducible. For lenders, the implication is stark: a clean, well-organized loan package is no longer a hallmark of authenticity.

Why CRE Is Particularly Exposed

Commercial real estate lending is structurally more vulnerable than other credit asset classes. Three characteristics compound the risk.

First, the sector’s dependence on borrower-controlled reporting creates an information environment with few natural checkpoints. Information is rarely public, so there is a heavy reliance on self-reporting. Unlike public corporate credit, where audited financials, regulatory filings, enhanced governance and market pricing provide independent verification, CRE operating data flows from borrower to lender with limited validation.

Second, there is a secondary “backstop” reliance of the creditworthiness of sponsors or guarantors. That reliance is often insufficient or misplaced when fraud is being perpetrated at the guarantor-level as well.

Third, every CRE deal is bespoke. Customized terms, property-specific assumptions and negotiated structures resist the standardized verification frameworks present in consumer credit. Each transaction is evaluated on its own terms – and those terms are defined largely by documentation the borrower provides.

The Moments of Greatest Vulnerability

For credit investors and asset managers, understanding when the risk peaks is as important as understanding how it operates. The highest-risk moments are not necessarily at origination. They occur at all points in the investment cycle but are elevated when borrowers face decisions that hinge on reported financial performance – and when the consequences of those decisions are largest.

Annual self-reporting by borrowers and guarantors is susceptible to fraud. The very safety valve to support the integrity of the credit can be spoofed Similarly, on construction draws and earnouts, lenders can be tricked by falsified invoices and other documentation.

Economic downturns and negative trends can be powerful incentives to provide deepfake documentation to lenders.

Current market conditions sharpen those incentives considerably. The rise of AI has pushed private capital investment toward hard assets with low obsolescence, or “HALO” trades like CRE, at a time when significant volumes of CRE debt are maturing into a higher-rate environment, causing borrowers to face refinancing gaps, equity shortfalls, and lender pressure.⁵ AI dramatically lowers the barrier to producing documentation that supports a more favorable narrative – precisely when lenders depend most on the integrity of that data to make consequential credit decisions.

Three Pillars of AI-Enhanced Detection

The same technology reshaping fraud execution is also enabling a fundamentally more powerful detection architecture. The most effective approaches emerging across institutional credit platforms rest on three complementary pillars.

Portfolio-level financial anomaly detection. Machine learning models trained on large datasets of comparable properties can identify financial patterns that deviate from how real assets typically perform – operating expense ratios inconsistent with property type and geography, rent roll structures that lack the organic variability found in genuine tenancies, or sudden NOI improvements preceding refinancing events. These signals are often invisible at the individual deal level but become statistically unmistakable when properties are benchmarked against thousands of comparable assets.
Network and relationship analysis. Conducting deep-dive due diligence can help identify and profile individual and organization counterparties, business affiliations and ownership information based on public records, open sources and online footprints (current and historical). Using graph analytics to map the web of entities, guarantors, brokers and intermediaries across an entire portfolio often reveals patterns that deal-level diligence would miss. Some examples include the same intermediary appearing repeatedly across unrelated transactions, clusters of borrowers linked through shared beneficial owners, or coordinated filing patterns across entities that suggest organized misrepresentation. Conversely, the lack of information about an entity can be identified and flagged.
Document and media forensics. AI-driven tools can examine file metadata for inconsistencies in creation history, identify template reuse across submissions, detect pixel-level manipulation artifacts in scanned documents, and evaluate written narratives for linguistic markers – such as statistical predictability and uniformity in sentence structure – that may indicate machine-generated text in borrower explanations or appraisal commentary.

No single pillar is sufficient on its own. The strategic value lies in combining all three at scale, creating a surveillance architecture that captures and monitors borrower behavior and property performance continuously rather than at discrete underwriting.

The Strategic Imperative: From Underwriting to Surveillance

The traditional underwriting model – intensive due diligence at origination, periodic covenant checks and reactive investigation when problems surface – was designed for an environment in which fabricating convincing documentation required meaningful effort and skill. That environment no longer exists.

Credit committees and asset management teams are less likely to see a borrower’s loan package that isn’t internally consistent. What should worry them more is whether the reported performance is consistent with what comparable assets, prevailing market conditions and entity behavior patterns would independently predict. Answering that question requires a shift from static, deal-level underwriting to continuous, portfolio-level surveillance – monitoring not just financial metrics but the integrity and plausibility of the data itself.

Institutions that build or access these portfolio-level surveillance capabilities are better-positioned to identify deteriorating credits earlier, and to obtain a greater information advantage to negotiate workouts and mitigate losses that come from discovering fabricated performance data long after the critical decision point has passed. Those that continue to rely on the traditional deal-level model will likely find themselves increasingly blindsided by documentation that appears complete but tells them less and less about the truth underneath.

The steep reduction of friction in producing CRE documentation caused by AI adoption has allowed potentially fraudulent deals to hide more effectively at a time when the cost of fraud can be higher than ever. And in a market defined by stress, refinancing pressure and acute information asymmetry, the ability to distinguish authentic borrower performance from manufactured narratives is fast becoming the most consequential edge in commercial real estate credit.

Key AI Concepts in Fraud Detection

Anomaly Detection	Identifies financial or operational patterns that deviate from expected norms across comparable assets (e.g., expense ratios, rent distributions, NOI trends).
Network (Graph) Analysis	Maps relationships between borrowers, guarantors, and entities to uncover hidden connections, shared ownership, or coordinated activity.
Document and Metadata Forensics	Examines file attributes – timestamps, authorship, and template structure – to detect manipulation or reuse across submissions.
Perplexity (Text Predictability)	Measures how statistically predictable a sequence of text is. AI-generated content often exhibits lower perplexity, appearing smoother and more uniform than human writing.
Burstiness (Linguistic Variability)	Captures variation in sentence structure and word usage. Human writing tends to be more uneven; AI-generated text is often more consistent.
Cross-Document Consistency	Tests whether data reconciles across an entire package. Perfect consistency, while historically a positive signal, may itself warrant scrutiny.