The Legal AI 'Dry Run' Problem: Why Law Firms Are Simulating Deals and Disputes With AI Before They Happen — and What That Changes About How Lawyers Prepare

The pre-matter simulation is becoming the most consequential — and least discussed — shift in how sophisticated legal practices actually work. Before a deal closes, before discovery opens, before the first demand letter goes out, a growing number of BigLaw and mid-market firms are running AI-powered dry runs: feeding deal structures into simulation engines, modeling how opposing counsel will attack a litigation theory, war-gaming counterparty positions in contract negotiations. The matter hasn't started. The billing clock may not even be running. And already, lawyers know more than they ever did at this stage.

This changes everything about preparation. It also raises professional responsibility questions the bar has not seriously reckoned with.

What the Simulations Actually Look Like

In M&A due diligence, firms are using tools built on top of large language models — some proprietary, some adapted from platforms like Harvey or ContractPodAi — to model the risk profile of a target company before formal diligence even commences. Feed in publicly available filings, industry litigation patterns, the acquiring client's stated deal rationale, and the simulation surfaces the most likely legal vulnerabilities: regulatory exposure under FTC second requests, representations and warranty clauses that target-side counsel will resist, indemnification structures that comparable deals have seen blown up in arbitration.

Skadden and Kirkland are not publicly announcing these capabilities. But practitioners at both firms have described, in conference settings and legal tech forums, running what amount to adversarial AI simulations during deal origination — before engagement letters are signed, in some cases.

In litigation, the use case is more viscerally obvious. Pre-trial strategy simulations model how opposing counsel will frame a summary judgment motion, which of your expert witnesses will be attacked under Daubert v. Merrell Dow Pharmaceuticals (1993), and what a judge with a particular published record is likely to do with a novel evidentiary issue. Tools like Lexis+ AI and Casetext's litigation analytics layer have been used this way for two years now, but the 2025 generation of simulation capability goes further — modeling entire pre-trial sequences, not just individual motions.

In contract negotiation, particularly in complex technology and licensing deals, firms are pre-loading known counterparty positions — often reconstructed from prior deal tombstones, public filings, and litigation history — to simulate how the other side's counsel will redline a master services agreement or resist a limitation of liability clause. This is preparation that used to take a senior associate three days of manual research and pattern-matching. The simulation runs overnight.

Why Partners Love It (And Why Associates Are Unsettled)

Partners love pre-matter simulation for an obvious reason: it compresses the timeline from engagement to confident strategic positioning. You walk into the kickoff call already knowing where the deal is likely to break, where opposing counsel has historically pushed back, and which of your own arguments has the weakest foundation. That is an extraordinary positional advantage, and clients are starting to expect it.

But for mid-level associates — the population that used to own the early research and strategy memo phase of a matter — the disruption is real and the professional development consequences are underappreciated. The simulation doesn't just do the research faster. It does the thinking faster: it surfaces the counterparty's probable arguments, ranks the litigation risks, proposes the opening negotiation position. The associate who would have spent two weeks developing that judgment, making mistakes, getting feedback, and building pattern recognition is now supervising an output they didn't generate and may not fully understand.

This is not a small problem. Legal judgment is apprenticeship-based. The simulation collapses the loop that creates lawyers. Firms need to think carefully about how associates are deliberately exposed to the error-correction and analytical reasoning that the AI has now automated — or they will produce a generation of lawyers who are excellent at validating AI output and poor at generating independent legal analysis when it matters.

The Professional Responsibility Questions Nobody Wants to Answer

Here is where the dry run problem gets genuinely hard. If an AI simulation accurately models opposing counsel's litigation strategy — and that accuracy derives in part from analyzing that counsel's prior work product, deposition patterns, and brief-writing style scraped from public court records — does that raise competency obligations, or something more uncomfortable?

The ABA's Model Rules of Professional Conduct don't contemplate this scenario. Rule 1.1 (competence) now has the 2012 amendment requiring lawyers to keep abreast of "relevant technology," but simulation output is not research and it is not work product in any traditional sense. It occupies an uncomfortable middle space.

More pressingly: who owns the simulation? If a firm runs a pre-matter AI dry run, generates a strategic analysis, and then — for whatever reason — does not take the engagement, what happens to that output? It may contain privileged analysis. It was generated using firm systems trained, in part, on prior client matters. The ownership and confidentiality questions are not resolved by any current bar guidance, and the In re: Legal Services AI Governance framework proposed by the Conference of Chief Justices in early 2026 doesn't reach pre-matter conduct.

Disclosure is the sharpest edge. If your simulation predicted, with reasonable specificity, that the other side would make a particular evidentiary challenge — and it did — and you had prepared a counter-strategy in advance — at what point does asymmetric simulation capability become an issue of fair dealing between counsel? It won't be framed that way for years. But it will be framed that way eventually.

Preparation Is No Longer the Same Thing It Was

The pre-matter simulation is not a research tool. It is a strategic intelligence tool that operates before the matter begins, collapses traditional preparation timelines, and reshapes the distribution of judgment inside a firm. Partners gain leverage. Associates lose formative experience. Clients benefit — until they're the counterparty whose moves were simulated.

The bar needs to catch up. Firms need to be honest about what this technology is doing to how lawyers develop. And practitioners who are using these tools — which is most of you, or soon will be — need to think carefully about what you owe the process you are running ahead of.