The largest plaintiff firm in America has unveiled MX2, a proprietary AI platform it says has been running at scale for years, alongside a minimum $1 billion technology commitment over the next decade. The proportions of that bet, not the headline number, are what BigLaw should be reading twice.
There is a particular tone that law firms adopt when announcing artificial intelligence spending. It is the tone of a firm that has just discovered fire, is very excited about fire, and would like you to know that its fire is proprietary.
Morgan & Morgan’s announcement this week has some of that. It also has something most of the genre lacks, which is a functioning product with a user count attached and a founder willing to say out loud what everyone in BigLaw says only at conferences after the second drink.
The firm has formally unveiled MX2, an in house AI platform it says now has close to 5,000 monthly active users, and committed to spending at least $1 billion on technology and AI over the next ten years. Close to $500 million of that will be spent in the next five. It builds, the firm says, on $300 million already spent over the past five years.
On the raw number, that puts Morgan & Morgan behind Kirkland & Ellis, which committed $500 million over three to four years to build its own platform, beginning with around $100 million in 2026.
The Key Percentage
Kirkland reported revenue of $10.6 billion for 2025. A $100 million tranche in 2026 is therefore somewhere around one per cent of turnover. Serious money, seriously deployed, but roughly one cent in the dollar.
Morgan & Morgan does not publish accounts. Artificial Lawyer, working from a Reuters report in June putting the firm’s revenues at $2.4 billion, calculated that $100 million a year represents around four per cent of total annual income.
The comparison is stark – four per cent against one per cent. A plaintiff firm is committing, proportionally, about four times what the highest grossing law firm on earth is committing. That is the story. Everything else is packaging.
It is also, and this is the part that should hold a managing partner’s attention, a considerably easier decision for Morgan & Morgan to make.
Why Contingency Fees Change the Arithmetic
John Morgan’s framing is blunt to the point of rudeness about the rest of the profession. Firms billing by the hour to draft, review and read through documents are, in his account, the practices AI will replace almost entirely.
Trial lawyers, he argues, are a different proposition. “No robot is walking into a courtroom to deliver a closing argument.”
You can dismiss that as a plaintiff lawyer’s showmanship, and some of it is. The underlying economics are harder to wave away.
A contingency firm does not have a realisation problem. If MX2 builds a medical chronology in forty minutes rather than four hours, nobody has to have an awkward conversation with a client about a write down, and no partner has to explain to the compensation committee why the matter came in under budget.
The saving falls directly to the firm. Faster demands, faster resolutions and more cases handled per lawyer are all pure margin.
BigLaw’s position is structurally worse, which is why its AI announcements read the way they do. Kirkland chair Jon Ballis (Pictured) has been notably candid about it, framing the firm’s platform as a move towards value based pricing rather than a faster way to bill hours, an approach LawFuel examined at length when the Kirkland spend was announced.
But that is a firm arguing itself out of a business model that currently produces record profits per partner. Morgan & Morgan never had to make the argument. It was already paid on outcomes.
The uncomfortable summary for hourly firms: the plaintiff bar can adopt AI enthusiastically because efficiency and revenue point in the same direction. For the defence side, they point in opposite ones.
The MX2 History
MX2 is not making its debut. There was an earlier iteration, MX2.law, which was already in use inside the firm, and in February 2025 it produced eight non existent authorities in motions in limine filed in a Wyoming product liability case against Walmart.
Judge Kelly H. Rankin fined the associate who used the tool $3,000 and revoked his pro hac vice admission, with the supervising attorney and local counsel fined $1,000 each.
LawFuel covered the sanctions and what they meant for signing partners at the time.
The point is not that Morgan & Morgan should be permanently defined by it. Most large firms have an AI incident they would prefer not to discuss, and the ones that do not are usually the ones that have not deployed anything.
So what changed between the version that fabricated case law and the version now handling hundreds of thousands of demand letters?
Verification architecture is the whole ballgame in a plaintiff practice. Medical record extraction, chronology building and automated demand generation are exactly the workflows where a confident hallucination becomes a client’s damages claim rather than a citation error. The firm says MX2 spots patterns and sets benchmarks across cases. The interesting disclosure, when someone gets it, will be how MX2 is prevented from asserting things that are not there.
Build, Buy, or Something in Between
Morgan & Morgan’s build sits inside a market that has already fragmented into distinct strategies.
| Firm | Approach | Stated commitment | What it signals |
|---|---|---|---|
| Morgan & Morgan | Full in house build, engineers working alongside trial lawyers, embedded in Litify | At least $1bn over 10 years, close to $500m over 5 | Contingency economics make efficiency pure margin |
| Kirkland & Ellis | Proprietary platform encoding firm knowledge, vendor tools as replaceable layers | $500m over 3 to 4 years, circa $100m in 2026 | Scale firms want an owned asset competitors cannot licence |
| Willkie Farr | Collaboration with OpenAI, firmwide ChatGPT Enterprise, in house build shop | Not disclosed | Frontier lab partnership rather than solo engineering |
| Latham & Watkins | Ecosystem approach, AI as a practice area, mandatory associate training | Not disclosed | Monetise AI advice as well as consume AI tools |
| A&O Shearman, Clifford Chance, Mishcon de Reya | Vendor partnerships and internally branded platforms | Not disclosed | Middle path, own the interface, rent the intelligence |
Meanwhile the vendors are integrating vertically underneath all of them. Harvey has built a custom legal model called Tenet, and Thomson Reuters has launched its own proprietary model, Thomson, developed in house with a reported $40 million investment and first deployed in CoCounsel Legal.
Which raises the obvious question about what “proprietary” is actually worth. Almost every firm platform still runs on somebody else’s foundation model. The defensible asset is not the model. It is the data, the workflow encoding and the institutional judgment layered on top. Morgan & Morgan’s version of that asset is forty years of injury litigation outcomes. Kirkland’s is how its partners actually run deals. Neither is buyable.
The 2027 Move
Buried near the end of the announcement is the line that matters most commercially. Morgan & Morgan intends to make MX2 available to other law firms, by invitation, by the end of 2027, and it specifically names corporate and transactional practices as the target.
It seems somewhat ironic, but the largest plaintiff firm in America proposes to sell software to the defence bar, having first told the defence bar that AI will replace most of what it does.
There is also a second reading, and it is not cynical so much as financially literate. Reuters reported in June that the firm was weighing a minority stake sale that could raise more than $1 billion, along with a longer term IPO. A law firm shopping equity is valued as a law firm. A law firm with a proprietary AI platform, a licensing roadmap and a decade long capital commitment starts to look like something investors price differently.
What to watch
- Verification disclosures. The credibility of MX2 in litigation workflows rests on what sits between generation and filing. Nobody has published that yet.
- The 2027 licensing terms. Invitation only access for corporate firms is either a serious product play or a positioning line. The pricing will tell you which.
- Whether the proportional spend spreads. If mid sized contingency and volume practices read the four per cent figure and act on it, the AI investment story stops being a BigLaw story.
- Any stake sale. If external capital arrives, every claim in this announcement becomes a disclosure document rather than a press release.
Figures on spend, user numbers and results in this article are as stated by the firm and are not independently audited. Morgan & Morgan does not publish revenue. Revenue estimates are as reported by Reuters.
