If you're reading this, you've probably already decided the question isn't whether your firm adopts AI. It's which AI, and how. Increasingly, firms are arriving at the same answer to the first half of that question: Claude. This post is about the second half.
Why Claude, specifically
In May 2026, Anthropic launched Claude for Legal — a formal, comprehensive push into the legal industry, including 12 new practice-area plugins covering everything from commercial counsel to litigation support, and more than 20 new integrations with tools firms already rely on. It wasn't a quiet feature release. Adopters named alongside the launch included Freshfields, Quinn Emanuel Urquhart & Sullivan, Holland & Knight, and Crosby Legal, all already running Claude on live matters.
That matters for a simple reason: it means the frontier model itself is being built, in partnership with major firms, specifically for legal work, not retrofitted from a general-purpose tool after the fact.
Quinn Emanuel's Christopher Kercher, the firm's Head of AI & Data Analytics, has described building the firm's litigation platform directly on Claude despite having no formal coding background. Freshfields co-head Gerrit Beckhaus has called Claude's capabilities essential to the firm's proprietary AI tools, with the firm now co-developing agentic workflows with Anthropic directly.
That's the environment North builds in: not a startup betting on which model might win, but the frontier lab's own legal push, with North as the team that installs it firm-wide.
Wrapper vs. native: what the difference actually means
"Wrapper" gets thrown around loosely, so it's worth being precise about what it means technically.
A wrapper tool sits on top of a model it doesn't control. When the underlying model improves, the wrapper has to catch up — rebuild prompts, retest workflows, sometimes wait months for feature parity. The tool is only ever as current as its last update cycle, and firms are locked into whatever architecture the vendor chose two or three product cycles ago.
Building directly on Claude means a firm's system improves as Claude improves, without a middle layer slowing that down. Two capabilities matter especially for legal work. The first is context length: through the API and certain enterprise tiers, Claude's extended context window covers roughly 1,500 to 2,000 pages in a single session — the tier North deploys on — which for legal work isn't an abstract spec, it's the difference between summarizing one contract at a time and holding an entire diligence binder, a full discovery production, or an entire deal's document set in a single working session, with nothing paraphrased or dropped along the way. The second is writing quality. Lawyers who've worked with Claude consistently point to how compatible its writing style is with how legal documents actually read, a real factor when the output is a redline, a client letter, or a brief, not just a research summary.
Use cases, by role
Claude's legal deployment isn't one tool doing one thing. Partners tend to use it for matter oversight, client communication drafting, and cross-matter research synthesis. Adoption tends to run deepest among associates: senior counsel at Jasper described switching almost entirely to Claude for redlining and contract review once the Word integration rolled out, citing how naturally it handled reviewer comments and off-market clause flagging without leaving the document. Paralegals and clerks use it further downstream still — intake processing, conflict checks, deadline calendaring, matter briefs.
North's own skill library reflects this range directly: closing binders, signature packets, title review, NDA review, entity formation, deadline calendars, client letters, diligence indexes, cap tables, intake, conflict checks, and matter briefs, each one a governed, reusable skill rather than a one-off prompt someone wrote and never revisited.
Security and governance: the objections that actually come up
Every firm evaluating Claude asks some version of the same three questions.
Who owns our data? This is not a hypothetical concern. Firms that have tried to leave other AI providers have reported delays, withheld access, and steep extraction fees just to get their own data back, averaging roughly £12,888 in the UK and A$24,861 in Australia in reported cases. Claude does not train on firm data on its enterprise tiers, and work stays behind the firm's own access controls.
Is the output trustworthy? This is the uncomfortable one. Courts have logged more than 1,800 decisions worldwide involving fabricated AI-generated citations as of August 2026, and even Anthropic itself was not immune: one of its own outside lawyers was forced to apologize after a hallucinated citation slipped into a filing. The honest answer isn't that hallucination risk disappears with a better model. It's that governed deployment — citations tied to source documents, skills built and reviewed for specific practice areas, oversight instead of an open chat window — is what actually reduces it. That governance is the point of North OS, not an add-on to it.
What happens to adoption after the sales call ends? This is where most legal AI tools fail quietly. A license gets purchased, a training session happens, and six months later usage has dwindled to whoever was already inclined to use it. Across North's own deployments, firms relying on self-serve tools before working with North saw real adoption settle around 9%.
Implementation: what actually changes when North installs Claude
The technology is only half the story. The other half is how it gets into the firm.
North Deploy sends Forward Deployed Engineers on-site, not a remote onboarding call, but a team that learns exactly where a firm's hours go, matter by matter, before customizing anything. That tailoring — house style, existing templates, specific practice mix — is what turns a capable model into a system a firm actually adopts, rather than a tool a few people try once.
The result: across North's own deployments, firms going through structured, on-site implementation see adoption around 80%, against roughly 9% for firms left to adopt tools on their own.
Proof, not just a pitch
According to North, Hirschen Singer & Epstein LLP, a New York firm, had tried other legal AI tools before working with North. They sat unused. After North installed Claude across the firm, adoption was firm-wide within weeks.
That's the pattern North is built around: not whether Claude is capable enough for legal work (Anthropic's own legal push, and the firms building directly on it, have settled that question) but whether a firm can actually get the whole building using it. That's the problem North solves.
Where to start
If your firm is evaluating Claude, the real question isn't the model. It's who installs it, how it gets customized to your practice, and whether the whole firm ends up using it or just the handful of people who were already curious.
A 30-minute call is enough to map one real workflow and show exactly where it fits.
Sources: Reuters (May 2026); Law.com Legal Tech News (May 2026); Fortune (May 2026); LawSites/Robert Ambrogi (May 2026); Bloomberg Law (May 2025); GC AI (July 2026); Artificial Lawyer (April–May 2026); Clio State of Legal Tech Report, UK & Australia; AI Hallucination Cases Database, Damien Charlotin.