TL;DR
Google has built a legal tier into Gemini Enterprise, its business AI platform, adding connectors to legal data systems and a set of agents pitched as able to run specialist and administrative tasks without close supervision. Freshfields is among the firms working with Google on it. A financial services tier launched the same day.
A hyperscaler going direct
Until recently, selling AI into law meant selling through the legal technology vendors. This is Alphabet approaching firms itself. Gemini Enterprise opened to business customers last October, and Google Cloud has become one of the group’s quickest-growing lines; the legal tier extends that reach into a profession with unusually strict confidentiality obligations, which is why the announcement leads on keeping client data contained.
The timing is pointed. Thomson Reuters — which owns both Westlaw and Reuters itself — put out a proprietary model called Thomson 1.0 on Monday, built on open-source foundations and trained on decades of its own legal research material. Google’s platform will nonetheless connect to Thomson Reuters systems, as well as to Harvey and Legora. Anthropic has been expanding into the same vertical. Deloitte reported this year that firms are spending heavily while the change remains at an early stage.
What it means for UK firms
Freshfields sits alongside Weil Gotshal, Williams & Connolly and Cleary Gottlieb in the collaborating group — a London-headquartered magic circle firm helping shape a product that will land on its competitors’ desks too. Google Cloud product director Amrita Mehrok framed the flexibility as model choice: firms pick from the models embedded in the platform rather than being locked to one.
That reframes a question mid-sized UK practices have been circling all year. Building an in-house platform is a serious commitment; buying from a legal tech vendor has been the default. A hyperscaler offering an agent layer directly makes the middle path cheaper and the build case harder to justify.
Looking forward
Agents that act “without significant human oversight” will meet a profession where supervision is a regulatory duty, not a preference. Yesterday’s Nuffield-funded review found justice AI is being measured on speed rather than on outcomes — the same gap applies here. UK firms should expect their insurers and the SRA to ask what the agents did before their clients do.