/>iThis article is a slightly modified version of my weekly newsletter, Attorney Intelligence.
Last week, the legal AI world was buzzing about Harvey’s much-anticipated integration with LexisNexis. Adding Lexis’ complete repository of legal case data promised to be a big step forward in functionality for Harvey’s platform.
Unfortunately, the end result might not be quite what it was hyped up to be. According to sources who've seen demos of the new integration, Harvey’s access to the Lexis data set is much more limited than many expected.
In this article, we’ll cover:
- Why legal AI companies need data access
- How Harvey’s Lexis integration actually works
- The incentives of the various players
- How this problem ultimately gets resolved
Let's dive in.
Keys to the Kingdom
Legal AI companies face a serious problem: lawyers won't trust AI that can't cite real cases.
Without access to primary sources, Harvey, Legora, and every other legal AI platform will always be fundamentally limited. These platforms have beautiful interfaces, integrate with Word, and enable lawyers to work with internal documents and publicly available data sources like EDGAR. But to achieve their ambitions – revolutionizing the practice of law – it’s essential that they acquire complete, robust, and reliable repositories of case data.
The real power in legal research lies with three companies: Thomson Reuters (Westlaw), LexisNexis, and vLex. They control the comprehensive databases of case law, statutes, regulations, and — crucially — the editorial enhancements like headnotes and citations that make legal research actually work.
vLex was recently taken off the table in their $1B acquisition by Clio.
This is why Harvey's partnership with Lexis generated so much excitement. Finally, a leading AI platform would have access to one of the remaining "big two” legal databases.
The end result, while a step in the right direction for Harvey, is far from the silver bullet many hoped it would be.
What the Lexis Integration Actually Does
In a nutshell, Harvey’s Lexis integration works as follows:
- You type a legal research query into Harvey
- Harvey forwards your query to Lexis AI
- Lexis AI searches Lexis' database and generates an answer
- Harvey displays that answer in its interface
This does offer some immediate benefits: Harvey users can now access Lexis data without switching platforms, making it easier to incorporate Lexis search data into broader workflows.
The problem is, Harvey has no access to the underlying data. It can't train on Lexis' corpus. It can't even see how Lexis generates its answers. Harvey’s interface has essentially become a nice wrapper for Lexis’ own AI product.
The Ceiling on “Lexis Inside”
To put it simply, if Harvey is stuck sipping through the straw of Lexis AI, its outputs can never be better than what Lexis can give it. For a company that positions itself as being at the cutting edge, this is a very meaningful limitation.
With a more comprehensive integration, we'd see:
- Unified reasoning across Harvey's AI and Lexis' data
- The ability to train or fine-tune custom models on case law
- Seamless blending of internal knowledge and public law
- A single AI brain with access to everything
Unfortunately, what materialized looks more like an API forwarding service.
Don’t get me wrong, this is still a positive for Harvey’s product and will likely add value for their users. But there is nothing fundamentally new here: Lexis customers already had access to this information; it’s merely available in a different interface. And Harvey’s restricted access limits their ability to continue building on top of the Lexis data.
The Siri Lesson
Last year, Apple realized it had an AI problem. Google and OpenAI had jumped out ahead, and while progress on their internal AI models was progressing slowly, the public demanded a solution – fast.
In December 2024, at the launch of Apple Intelligence, Apple announced it had integrated ChatGPT into Siri. Now, users would be able to leverage the cutting-edge intelligence of ChatGPT directly from their iPhones.
Anyone with an iPhone probably knows how this story ends.
In reality, Siri became a forwarding service for the ChatGPT API, with very little “integration” beyond that. The resulting product was hardly more convenient (and in many ways worse) than going directly to ChatGPT.
That doesn’t mean that Harvey’s Lexis integration will follow the same fate. But it is a cautionary tale in how these partnerships can go wrong if they aren’t thoughtfully executed and deeply integrated.
Why the Vault Stays Shut
Why won't Lexis give Harvey full data access? Simple: their entire business relies on the value of that data.
LLMs are commoditizing fast. Anyone with a credit card can start building on top of them.
But legal databases? Those take decades to build. The headnotes, Shepard's citations, and editorial analysis represent millions of hours of attorney work that can't be replicated overnight. And law firms have extremely high standards for the quality and completeness of the data. Even 99% market coverage is insufficient – that remaining 1% can cause disastrous outcomes. Something very close to “absolute completeness” is required.
Lexis simply doesn’t have adequate skin in the game to throw open the gates to their data set. Their equity investment in Harvey is small relative to Harvey’s massive $5B valuation. Meanwhile, Lexis’ own business would be put at risk by giving Harvey full access to their data. Once the genie is out of the bottle, there’s no putting it back.
So – at least as of today – Lexis has chosen to keep the vault locked: no corpus access, no training data, no editorial logic.
How This Gets Solved
There’s an incredible amount of pressure (and dollars) driving Harvey, Legora, and the other legal AI companies to acquire case data.
There are only two realistic solutions to solving this problem:
Option 1: Buy your way in. The cheapest option, vLex, was just taken off the table by Clio. Since Lexis and Westlaw are too large to acquire, this probably involves a very large, expensive partnership that ties the financial fates of the two companies. It’s not clear whether the financial incentives exist for this type of deal to get done.
Option 2: Build your own. There are many smaller vendors who have some portion of the total data. An enterprising startup with lots of time and money could buy or license a few of these and cobble together the rest. Harvey recently announced a data licensing deal with Wolters Kluwer that looks like a step in this direction. This is also the path I’d expect Legora to pursue.
This is a steep product and business challenge. Whoever can solve it first will have a meaningful edge in the legal AI wars.
All of the views and opinions expressed in this article are those of the author and not necessarily those of The National Law Review.
