Legal AI – what is the end game?

While the headlines have backed off from generative AI (GenAI) signalling the end of humanity, what will it ultimately mean for lawyers and justice? If an AI model selects the most important facts for a court case, it is surely determining the case’s direction and potentially its outcome. As the options for applying GenAI to legal continue to evolve and expand, can benchmarks help firms find the best fit and leverage their tech budgets? And the latest fallout from the Weil, Gotshal & Manges collapse reinforces the fact that data ranks alongside talent as a law firm’s most valuable asset. 

AI-assisted justice?

At Legal Geek last month, Sir Geoffrey Vos, speaking in his final two weeks as master of the rolls, stepped back from years of AI evangelism to express concerns about the end game. What happens when AI, already used by the judiciary, serves to distance judges from the cases they decide? 

‘We may be on the edge of a slippery slope,’ observed Sir Geoffrey. He believes a serious international debate is needed to establish, for example, whether AI-assisted decisions made in one jurisdiction are enforceable in others. He referenced the Standing International Forum of Commercial Courts (SIFoCC)’s AI project. This ‘is attempting to distil good practice in the use of AI in commercial [courts] decision-making in 71 jurisdictions… to uncover uncontroversial use cases for AI, and to say where caution is required because of the risk of inadvertently reducing judicial involvement in decision-making… SIFoCC’s AI project will likely end up developing model rules for the use of AI’. 

While well-intentioned, this raises questions about how much difference introducing model rules will make, when AI is already widely used to help judges resolve cases more quickly and reduce court backlogs.

Bidding for data

'Data acquired from Weil could be combined with Scissero’s existing resources to accelerate the development of a new, fully agentic product that would operate without lawyers routinely involved in the process'

Mathias Strasser, Scissero

In relation to a different end game, while last week’s demise of US white shoe law firm Weil Gotshal was unrelated to tech or AI – it was triggered by a failed merger followed by a talent exodus – three AI-native law firms have expressed interest in the firm’s tech and data assets and potentially its remaining talent. 

New York-based AI-native law firm Crosby was the first to confirm it had submitted a letter of intent to buy Weil assets, including its trade mark, its brand and some unspecified internal data. Mathias Strasser, CEO of London-based AI-native law firm Scissero, which acquired Robin AI’s managed services business last December, then confirmed submitting a formal expression of interest in acquiring Weil’s tech, IP and data assets. This included DealVision360, its proprietary AI-enabled private deal terms database, which provides data-driven insights and market benchmarking capabilities drawn from the firm’s transactions data. Also in London, Vetta, a recently launched AI-native law firm specialising in private equity, expressed interest in Weil’s legal precedents and other knowledge and tech resources. All three are keen to recruit Weil lawyers.

This new direction demonstrates that data is not just valuable as part of an M&A deal; data and tech have standalone value. Strasser was quoted as saying that ‘data acquired from Weil could be combined with Scissero’s existing resources to accelerate the development of a new, fully agentic product that would operate without lawyers routinely involved in the process’. 

Crosby is also looking at agentic lawyering, having recently discussed buying professional indemnity insurance for its AI agents. In legal, AI is a differentiator only when combined with proprietary data, and this is the first public bidding war for legal proprietary data.  

Intake trilemma

Are great minds thinking alike or is legal AI running out of innovative ideas? On 13 and 14 October, respectively, Legora and Harvey launched agentic AI products called Intake for in-house legal. Both are described as a digital front door for internal business requests to legal departments. By leveraging AI agents to automate routine tasks, gather required information and create initial drafts, Harvey and Legora aim to reduce administrative work for in-house legal teams, boosting efficiency and freeing up lawyers for more complex work.

While there is a slight difference in how initial requests come through, the principle is basically the same. Requests are submitted to Legora’s Intake via Slack, and an AI agent asks follow-up questions, assigns tasks and uses the legal team’s playbooks to prepare initial work for human review. 

Harvey’s Intake captures requests from multiple channels – email, Slack, Jira and so on – and directs them to the appropriate AI agent or lawyer, and the response goes back to where the enquiry originated. 

UK scale-up Wordsmith, an agentic platform for corporate legal departments, launched its own Intake product in August, an agentic workflow that also applies legal’s playbook to resolve routine requests arriving through the firm’s existing channels.

All three products effectively triage requests and automatically apply existing playbooks, allowing lawyers to be involved later in the process. 

Go big (tech) or open source?

Also in October, Microsoft launched Copilot for legal, becoming the fourth AI giant in five months to build out a legal vertical. But there are two key differences. First, nearly all law firms use Microsoft and Copilot has long been the gateway to legal AI for firms of all sizes, not just the ones that can afford to invest in Harvey and Legora. Second, Microsoft announced that Copilot for legal includes more than 20 purpose-built legal skills and partner plug-ins with leading legal tech and legal AI companies, which surely means that it does not anticipate replacing them.  

Another new direction for Big Tech’s incursion into legal relates to expertise. Anthropic has just recruited a conceptual reasoning fellowship for law for a four-month project working alongside its research team ‘to evaluate, improve, and apply Claude’s capabilities in their field’. Is this an extrapolation of Harvey’s recruitment of lawyers and legal innovation leaders to shape its products and engage the legal market? 

The alternative to the buy and/or build dilemma is to adapt open-source models. Mike OSS, an open-source, self-hostable equivalent of Legora or Harvey, created by former Latham & Watkins associate Will Chen, is the most popular. Mike OSS has 1.4k forks on GitHub (1,400 forks means that 1,400 users have copied the code to adapt it for their own purposes, to build on it, or to contribute to it via a pull request). However, deploying open-source models requires technical skills to install, configure and maintain them.

Allen Morgan, founder and CTO of legal software engineering and rules-based AI company Altien, has built a version of Mike OSS on Microsoft Azure. MikeOssAzure is specifically for law firms and provides the full version of the open-source platform ‘in a trusted Azure tenancy’, offering the resilience and security of the Microsoft cloud. Altien maintains the packaging and offers (paid) customisation and support. Morgan explains that because it deploys inside the firm’s own Azure environment, no operational (or confidential/privileged) data leaves the tenant. 

New benchmarks

Also at Legal Geek, Anna Guo and Elgar Weijtmans presented Legal Benchmarks, the first independent legal AI benchmarking initiative since early GenAI research by VALS, which was limited after leading vendors declined to participate. Guo and Weijtmans are crowdsourcing reviews from lawyers and other legal professionals (including, apparently, people who work for legal AI vendors) who use frontier AI models and legal AI platforms for specific legal processes. This is much more comprehensive, as lawyers use different combinations of models and wrappers, and analysis includes determining the best model weights to complete legal processes effectively while minimising token use. While Legal Benchmarks’ research is currently based on too few reviews to be statistically valid, it is a free resource and certainly one to watch.

Benchmarking can provide clarification to support procurement decisions, particularly for mid-market and smaller firms. Law firms are significantly increasing IT budgets and AI investments, but the legal tech stack is becoming more complex and cumbersome when more tools and systems are added, but none are swapped out. Is a rapidly evolving market for frontier and legal-specific AI driving adoption or standing in the way? And are connectors and integrators opportunities for differentiation or confusion?  

The rebranded legal tech unicorn Clio is taking a different direction, offering a cutting-edge AI version of the one-stop shop. CEO and founder Jack Newton announced multiple new products built in-house. Stephen Embry, reporting for Above the Law, wrote: ‘So who wins, the integrators or the unifiers?’ Right now, surely there is a market for both.