New foundations?
On 24 June, Perplexity launched Computer for Counsel, which connects its agentic platform Perplexity Computer to legal use cases. There are now five major foundational models directly targeting the legal market with legal AI tools and development partnerships: OpenAI, Anthropic, Microsoft, Palantir and now Perplexity. And while Google does not have legal-specific offerings, Anthropic is built on Google Cloud. Freshfields, the first Magic Circle firm to partner directly with Big Tech to build legal AI tools, has strategic development partnerships with both Google and Anthropic. Other firms are building specialist AI tools using general-purpose AI platforms: Kirland & Ellis partnered with software and data analytics giant Palantir to build its Fund Formation Engine and Fried Frank’s built its FundAssist on OpenAI models.
While there are discussions about legal AI shifting from integration to infrastructure, the tech giants and foundational AI models aren’t waiting for law firms to make AI part of their infrastructure. Microsoft provides the IT infrastructure for most law firms, many of which have rolled out Copilot, and the MCP (model context protocol) provides Claude (and other AI applications) with deeper systems integration that enhances their (agentic) capabilities, which in turn requires access to applications and data.
It follows that legal’s excitement about AI transformation is accompanied by an undercurrent of anxiety, because the sector that has long obsessed with predicting its own future is in the vanguard of disruptive technological change that is outside its control, with frontier AI already embedded in its tech stack. Which raises the obvious question: What does this mean for legal AI?
Legal AI promoted to senior associate
Legal AI is responding to the actual and perceived threat from frontier AI models by developing its own. Harvey is developing its own legal foundation models. On 17 June co-founder and president Gabe Pereyra posted on X that these would work alongside third-party models, with the goal of enabling law firms to build and own proprietary legal AI models (which some are already doing). In another post he explained that Harvey was working with multiple research partners on different aspects of fine-tuning models. On 22 June, Thomson Reuters announced early access to the next generation of CoCounsel Legal, built on Anthropic’s Claude Agent SDK (software development kit) and differentiated by what it describes as ‘fiduciary-grade AI’, i.e. grounded in authoritative, traceable content, as well as offering multiple integrations. And last week Thomson Reuters president and CEO Steve Hasker told US legal tech commentator Bob Ambrogi that Thomson Reuters’ proprietary large language model for legal, Thomson, is at an advanced stage of testing at Imperial College, London. Interestingly, both Harvey and Thomson Reuters describe their latest AI initiatives as the equivalent of a senior associate, so it looks like the new legal AI models are moving up the law firm hierarchy.
The AI arms race
When you think about the international element of the AI (arms) race, the strong interest among legal AI companies and larger law firms in building proprietary models and the increased prominence of open-source legal AI and vibe-coding make sense. On 13 June US authorities ordered Anthropic to suspend foreign nationals, including Anthropic employees, from using Claude Mythos 5 and its publicly available counterpart Claude Fable 5 because of national security concerns. Anthropic subsequently disabled Fable 5 and Mythos 5 for all their customers. Switching off access to the latest AI model fuelled the debate in the UK and Europe around international dependence on US technologies, including AI, with politicians calling for urgent investment into sovereign AI, which was a key theme at London Tech Week.
On 26 June, the US government allowed Anthropic to restore access to Claude Mythos 5 to a defined list of US organisations. Meanwhile, OpenAI also announced that it was complying with the US government’s request to limit the roll-out of its new GPT-5.6 models to a “small group of trusted partners”.
AI in court
At the Expert Witness Institute annual conference, Sir Geoffrey Vos, Master of the Rolls and Head of Civil Justice for England and Wales discussed AI’s growing role in litigation and said that ignoring AI is not a realistic option. However, he also acknowledged that it needs to be used appropriately with recognition of its limitations and called for an active debate on where the red lines should sit. While Sir Geoffrey is a well-known AI evangelist, perhaps given the continuing frequent appearance of AI hallucinations in court, the role of expert witnesses may become more important than ever.
Meanwhile, Garfield AI, the first SRA-regulated AI law firm, won its first case in Wandsworth County Court, in what it claims as the first trial won by an AI lawyer. The client, a freelancer recovering unpaid fees, used Garfield AI to prepare the documents for the case, which was argued by a barrister in a three-hour hearing. The client paid £400 in fees and was awarded £7,000.
Price hikes soon?
It looks like legal AI may follow another tech giant, Apple, and increase its prices. The cost of GenAI is rising because agentic AI’s ability to work autonomously is increasing token use. Harvey CEO Winston Weinberg told Sourcery that while the company grew 3x since last August (ARR up from $100m to $300m), its monthly token consumption increased 12x from 1 trillion in January to 12 trillion to 13 trillion in May.
Meanwhile Legora has introduced consumption-based pricing for its most capable product, Agent Pro.. Does this mean that instead of forcing legal away the billable hour, it is (kind of) being introduced into legal AI?
London calling
London feels like the capital of legal AI right now. You have only to check LinkedIn to see how many international legal tech leaders have visited London recently. This and the UK broadsheets’ recent focus on legal AI puts London front and centre. And now legal AI is all over my local football teams in west London. Last week, Legora entered a multi-year partnership with Chelsea FC to display its logo on the sleeves of its training kits. Legora is also used by Chelsea’s legal team. Earlier this year, Harvey became the official legal AI partner of Fulham FC, so the west London derby on 24 August will be effectively sponsored by legal AI!
Learn more about upcoming Legal Geek events on our events page.
Written by Joanna Goodman, tech journalist
Photo credit (Joanna): Sam Mardon
