Legal AI in your cloud platform is a different proposition from legal AI bought as a product, and Google Cloud made that move on 25 August 2026 with Gemini Enterprise for Legal. It arrives in preview rather than general release, carrying skills for contract review and redlining, playbook creation, regulatory horizon scanning, legal research and data subject access requests. Thomas Kurian, the chief executive of Google Cloud, put his own name to the announcement.

The early adopters named alongside it are Cleary Gottlieb, Freshfields, Weil and Williams and Connolly, and none of them resembles a firm of eight fee earners in Stockport. The part that reaches you sits further down the page, in the list of connectors. The product joins to Microsoft 365 and Google Workspace, to iManage and NetDocuments, to Docusign, Everlaw and RelativityOne, and to Harvey, Legora and Thomson Reuters HighQ. Legal AI stops being a place you go and becomes a layer over the systems you already run.

Legal AI stops being a place you go and becomes a layer over the systems you already run.

The decision loses its signature

When AI sits inside a legal research subscription, the choice to adopt it carries a date and a signature. Someone compared products, read the terms and signed an order form. When the same capability arrives in the cloud tenancy that already holds your email, your documents and your matter files, adoption happens the moment an administrator enables a feature, and nobody in the firm writes down a decision at all.

Give one person the job of knowing which AI features are switched on across every platform the firm pays for, and make it the person already accountable under your AI use policy rather than your IT supplier. The answer changes without notice, so the question belongs on a standing agenda rather than in an annual review.

What to test while it stays in preview

Google says client data, firm playbooks, custom agents and model outputs stay private to your organisation and are never used to train or fine-tune its foundation models. That is the right undertaking and it belongs in your contract rather than on a blog. The announcement also promises verifiable grounding and traceable citations, which is a claim to test on matters you have already closed, in the way the checking discipline sets out, before any of it touches live work.

Preview status matters for supervision as well. A product still being shaped by its launch customers behaves differently from one month to the next, and the checks your fee earners follow were written for the version in front of them at the time. The same caution belongs on a vendor that has built its own model.

Ask your IT supplier one question this week. Which AI features are enabled across the platforms we pay for, and who turned them on. If the answer takes longer than a day to produce, you have found the gap this announcement widens.

The announcement sits on the Google Cloud blog, open to read and asking nothing of you.

If you want to know what is already switched on across your firm's platforms before somebody else finds out, that is work we do: start with a conversation.