The Legal AI 'Choice of Law' Blind Spot: Why AI Contract Review Tools Default to the Wrong Governing Law Framework — and What That Costs in Cross-Border Deals
The pitch from every AI contract review vendor is essentially the same: faster review, fewer errors, consistent risk flagging. What they don't tell you is that the "consistency" baked into their models is overwhelmingly the consistency of Delaware corporate law, New York commercial practice, and...
The pitch from every AI contract review vendor is essentially the same: faster review, fewer errors, consistent risk flagging. What they don't tell you is that the "consistency" baked into their models is overwhelmingly the consistency of Delaware corporate law, New York commercial practice, and U.S. common law doctrine. If you're closing a joint venture with a Brazilian counterparty, acquiring a Gulf Cooperation Council target, or negotiating a supply agreement governed by French law, that consistency isn't a feature — it's a liability.
This isn't a fringe problem. It's a structural one, and it's quietly contaminating mid-market cross-border deal work at scale.
The Training Data Problem Is Worse Than Vendors Admit
AI contract review tools learn from contracts. The contracts available in bulk, in English, in machine-readable format, are predominantly U.S. agreements. Even tools trained on "global" datasets skew heavily toward common law drafting conventions because that's where the volume is. The result: models that treat New York or Delaware governing law as a kind of implicit default, even when the agreement in front of them explicitly says otherwise.
This surfaces in subtle ways that junior associates and legal ops teams aren't catching. The AI flags a clause as "market standard" or "low risk" without registering that the standard it's benchmarking against is inapplicable to the governing law the parties actually selected.
Where This Breaks in Practice
Force majeure is the most obvious casualty. Under New York law, force majeure clauses are interpreted narrowly and must explicitly enumerate triggering events — if it's not listed, it probably doesn't qualify, as courts reiterated exhaustively during COVID-era disputes (JN Contemporary Art LLC v. Phillips Auctioneers LLC, S.D.N.Y. 2020). AI tools trained on this jurisprudence will flag a broad, non-enumerated force majeure clause as vague or insufficiently protective.
But in French law, German law, or under the OHADA framework governing much of francophone West Africa, the analysis is entirely different. French law's imprévision doctrine — codified in the 2016 reform of the Civil Code under Article 1195 — allows courts to rebalance contracts when unforeseen circumstances make performance excessively onerous, entirely apart from what the force majeure clause says. An AI tool reading that broad clause through a New York lens may actively recommend tightening it in ways that strip the client of civil law protections they'd otherwise have by default. That's not risk mitigation. That's creating exposure.
Material adverse change (MAC) clauses present a parallel problem in M&A. U.S. practitioners have spent years calibrating MAC definitions against the Akorn v. Fresenius standard (Delaware Court of Chancery, 2018), where the court set an extraordinarily high bar for invoking MAC — a "durationally significant" effect on the target's business. AI tools trained on Delaware M&A contracts will evaluate MAC clauses against that framework, flagging narrow or company-specific MAC definitions as seller-friendly and recommending broader carve-outs.
In German M&A, the landscape is different. The concept of Störung der Geschäftsgrundlage under §313 BGB — disruption of the basis of the transaction — gives courts equitable tools to intervene in ways that don't map onto Delaware MAC analysis. A contract reviewed by an AI applying Akorn logic to a German-law acquisition agreement is being reviewed by a tool that doesn't understand the legal ground it's standing on.
Implied duties of good faith may be the most dangerous blind spot. U.S. common law has an implied covenant of good faith and fair dealing, but its scope is narrow — it generally can't override express contractual terms. AI tools calibrated on this understanding will treat an agreement's silence on good faith as unremarkable.
In civil law systems, good faith (bonne foi in France, buena fe in much of LATAM, Treu und Glauben in Germany) is a mandatory background obligation that courts apply broadly to performance, negotiation, and interpretation. In Gulf states with contracts governed by UAE law, Federal Law No. 5 of 1985 (the Civil Transactions Law) embeds good faith obligations throughout. An AI tool that doesn't weight these jurisdiction-specific implied duties will produce playbooks, risk summaries, and redline recommendations built on a foundation of doctrinal fiction.
What Sophisticated Practitioners Are Actually Doing
The international transactional lawyers I talk to have largely stopped trusting AI output on governing law issues without a human overlay layer. The practical workaround in use at serious international practices involves explicit jurisdiction tagging before AI review — essentially telling the tool "treat this as a French law agreement" — and then treating everything the AI flags as a hypothesis to be tested against local counsel input rather than a conclusion.
Some firms are building custom playbooks for their most common cross-border corridors: U.S.-Brazil, U.S.-UAE, U.S.-Germany. These playbooks function as jurisdiction-specific guardrails, but they're expensive to maintain and require genuine civil law expertise to build correctly. Kirkland and Latham have the bandwidth for this. Most mid-market firms don't.
Why Mid-Market Firms Haven't Noticed Yet
The answer is simple: the deals are still closing. Cross-border contracts with AI-generated memos full of U.S.-centric analysis still get signed. The problem doesn't announce itself at execution — it shows up in arbitration two years later, when a force majeure clause fails in a way the AI said was fine, or when a MAC invocation collapses because the governing law doctrine was never properly analyzed.
Mid-market firms doing cross-border work are essentially running an uncontrolled experiment on their clients' transactions, and the results won't be tabulated for years.
What Vendors Would Need to Do
The fix requires more than adding non-U.S. contracts to the training set. Vendors need jurisdiction-aware reasoning layers — models that identify governing law early in the review pipeline and modulate risk assessments accordingly. They need civil law partnerships with firms or academics who can provide doctrinal ground truth. And they need to stop marketing "global" capability based on multilingual text processing, which is a different thing entirely from multi-jurisdictional legal reasoning.
A handful of vendors — Luminance and Spellbook among them — have begun building jurisdiction-specific modules, but none have solved this comprehensively. The gap between what's marketed and what's functional remains substantial.
The Bottom Line
Cross-border transactional work is exactly the context where AI contract review should add the most value — reducing the friction of unfamiliar document conventions, accelerating due diligence across large data rooms, flagging structural inconsistencies. Instead, it's introducing a systematic bias toward U.S. common law analysis that sophisticated practitioners are compensating for manually and mid-market practitioners aren't catching at all.
The governing law clause is the interpretive key to every other provision in a cross-border contract. If your AI tool is reading past it, you're not doing AI-assisted review. You're doing U.S.-law review of a document that has never been subject to U.S. law.
That's not a minor calibration issue. That's a malpractice vector.