AI for NZ Law Firms: What Claude's Push Into Legal Actually Changes

TL;DR: On 12 May 2026, Anthropic launched a major expansion of Claude for Legal - twelve role-specific legal plugins, deep Microsoft 365 integration, and a model scoring 90.9% on the legal industry's hardest AI benchmark. For NZ law firms, the question stops being whether AI will reshape legal practice and becomes how exposed your firm is, and what your response actually looks like.

There is a particular kind of partner-meeting silence that follows a news article about AI in legal. The kind where everyone has read the headline on their phone over breakfast, but no one is quite sure whether to bring it up first - or whether they want the conversation to start.

This week's news is hard to leave on the breakfast table. On 12 May, Anthropic announced a significant expansion of Claude for Legal . Twelve role-specific plugins covering commercial, corporate, privacy, employment and AI governance work. More than twenty connectors into the systems law firms already use - DocuSign, Box, Thomson Reuters' Westlaw, the full Microsoft 365 suite. The underlying model, Claude Opus 4.7, scored 90.9% on Harvey's BigLaw Bench, the legal industry's most closely watched AI benchmark. Anthropic's own framing: you do not need to fine-tune a model to "give an engineer a legal degree" - the general-purpose model handles legal work well enough out of the box.

This is not another article telling you AI is coming for your job. It is an industry signal. For NZ law firms, the right response is not denial. It is also not panic. It is a practical, honest read of where you actually stand - and what to do about it.

What is Claude for Legal, and what makes this announcement different?

The Anthropic announcement is not a chatbot. It is a set of role-specific tools that plug into the systems law firms already use - DocuSign for execution, Box for matter files, Westlaw for research, Microsoft 365 for everything else. The plugins cover concrete practice-area workflows: a "commercial counsel" plugin for vendor agreements, NDA review and SaaS subscriptions with renewal tracking; M&A due diligence with disclosure schedule drafting; employment handbook drafting; and AI governance documentation.

The performance number matters more than it sounds. Claude Opus 4.7 scored 90.9% on Harvey's BigLaw Bench - the benchmark the legal industry has been quietly calibrating its expectations against. A year ago, "AI for law" mostly meant ChatGPT spitting out generic clauses you had to rebuild from scratch. Now it means a model that can hold its own on M&A due diligence and document review assignments designed to test real billable workflows. That is not a marketing claim. That is a benchmark with industry consensus behind it.

Anthropic itself is currently valued at US$380 billion, has reportedly been weighing funding offers north of US$900 billion, and may go public this year. This is not a side bet. And it is not alone. In the past two months, Harvey raised US$200 million at an US$11 billion valuation , and Legora raised US$600 million at US$5.6 billion . Thomson Reuters has extended its CoCounsel partnership with Anthropic . Microsoft has wedged Claude into the same Office suite your team has open right now. The competitive picture is no longer "will this happen." It is "it is happening - what now?"

Does this actually mean AI is replacing lawyers?

No. And it is worth saying clearly, because the honest answer matters more than the loud one.

What is actually happening is more interesting and more commercially important. AI is reshaping the economics of legal work - particularly the high-volume, document-heavy, judgement-low parts. Document review, due diligence sweeps, NDA mark-up, contract abstraction, first-draft clause generation. The work that has historically funded the bottom of the leverage pyramid is the work AI handles best.

It is also not yet without cost. The same week Anthropic announced its expansion, the public count of US lawyers caught filing AI-generated material with fabricated case law passed eighteen . California issued the country's first lawyer fine for ChatGPT-drafted submissions in late 2025. Federal judges have been flagged by Congress for using AI to draft rulings. The technology is genuinely capable. It is also genuinely embarrassing when used without the supervision a partner would apply to a junior associate.

But "lawyers are mishandling AI" is not the same as "AI should not be used." It is the same conversation any profession has when a powerful new tool arrives. The question is not whether to use it. The question is how to use it without negligence.

A confession that might irritate some lawyers

I have used AI to draft an NDA. I have used it to draft a sale and purchase agreement for business shareholding. I had reference documents. I had a base level of commercial knowledge to spot something obviously wrong. And I knew the risks I was accepting when I did it.

I am not a lawyer. I would not have used AI this way for anything contentious, anything with significant counterparty risk, or anything where the downside of getting it wrong outweighed the upside of moving fast. But the documents I produced were good enough for their purpose, and produced in a fraction of the time and cost a full legal engagement would have taken.

I am not pointing this out for effect. I am pointing it out because of what it means commercially. If I made that trade-off - accepting some risk for materially lower cost and time - your future clients will too. Some of them already are.

The question for a NZ law firm is not whether clients are starting to do this. It is whether your firm has a clear-eyed view of which parts of your current revenue are exposed to that trade-off, and which parts genuinely require the kind of judgement and accountability your clients pay for.

So what does an honest response actually look like for a NZ law firm?

This is where the AI Exposure Index that we use at Binary Refinery - what we call Monitor, Mitigate, Morph - gets practically useful. The framework is not diagnostic theatre. It is a way of separating which parts of a firm need active strategic work from which parts can be left to evolve at their own pace.

• Monitor (low exposure). Parts of the business where AI exposure is low and the right action is to stay informed. For most NZ specialist legal firms in 2026, very little of the practice falls cleanly into Monitor.
• Mitigate (material exposure). Parts of the business with real exposure that can be defended or hardened - usually by tightening processes, upskilling staff, or being more deliberate about what you charge for. Contentious work, complex commercial advice, court appearances, regulator engagement. The judgement-heavy work where AI augments but does not substitute.
• Morph (strategic transformation). Parts of the business where AI changes the underlying economics so completely that the existing model needs to be rethought. Document-heavy compliance work, standardised contract drafting, first-pass due diligence, basic legal research. The work that historically generated reliable revenue from associates and paralegals is the work where AI is closing the cost gap most aggressively.

Most NZ firms have all three bands in play simultaneously. The job is not to land on one number. The job is to know which parts of the firm sit in which band, and to make conscious decisions about each.

What the firms that get this right are already doing

The firms worth watching are the ones treating this as a strategic project rather than a technology purchase. In practice, that means a small number of commitments:

• Naming an internal AI lead with actual authority, not just an interested associate.
• Running a real exposure assessment of the firm's revenue mix - not a tick-box readiness audit, an honest one.
• Setting an AI policy that covers client confidentiality, professional obligation, and use of unvetted tools. The NZ Law Society's existing AI guidance is a starting point, not a finish line.
• Piloting AI in one or two practice areas with the supervision the work requires, and measuring real outputs - not vibes.
• Being transparent with clients about what is AI-assisted and what is not. The trust premium of legal practice is the whole asset.

That is a six-month workstream, not a Friday afternoon. But it is a workstream that produces a defensible firm position, not a reactive one.

If you are a partner reading this on your phone

You do not have to decide today whether Claude for Legal will be in your firm in six months. You do need to decide whether you have actually assessed your firm's exposure, or whether you have been hoping the conversation moves on.

It will not. AI tools for legal work have been around for years. This week, they got materially more serious. Your clients are watching. Some of them are already running the experiments.

If you want a structured way to work out where your firm actually sits - and what an honest, commercial response looks like for your practice areas, your client base, and your team - that is what the AI Disruption Risk Workshop does. Half a day, a clear AI Exposure Index score, and a practical view of where to focus first. If it is not valuable, you do not pay - that guarantee is the whole point.

Sources

1. Lucas Ropek, "The AI legal services industry is heating up. Anthropic is getting in on the action.", TechCrunch, 12 May 2026. https://techcrunch.com/2026/05/12/the-ai-legal-services-industry-is-heating-up-anthropic-is-getting-in-on-the-action/
2. "Claude For Legal Launches, May Reshape the Legal Tech World", Artificial Lawyer, 12 May 2026. https://www.artificiallawyer.com/2026/05/12/claude-for-legal-launches-may-reshape-the-legal-tech-world/
3. "Claude for Legal: What the industry needs to know", Legal IT Insider, 13 May 2026. https://legaltechnology.com/2026/05/13/claude-for-legal-what-the-industry-needs-to-know/
4. "Thomson Reuters and Anthropic Expand Partnership to Connect Claude with CoCounsel Legal", Thomson Reuters press release, May 2026. https://www.thomsonreuters.com/en/press-releases/2026/may/thomson-reuters-and-anthropic-expand-partnership-to-connect-claude-with-cocounsel-legal
5. "Anthropic Expands Push Into Legal Industry With New AI Tools", Bloomberg, 12 May 2026. https://www.bloomberg.com/news/articles/2026-05-12/anthropic-expands-push-into-legal-industry-with-new-ai-tools (paywalled; non-paywalled syndication at https://finance.yahoo.com/sectors/technology/articles/anthropic-expands-push-legal-industry-170010473.html)
6. "Even as hallucinations show up in legal filings, Big Law goes all in on AI with new Anthropic release", Fortune, 12 May 2026. https://fortune.com/2026/05/12/anthropic-legal-plug-in-release-claude-cowork-big-law/
7. "Claude Legal Solutions", Anthropic / Claude product page. https://claude.com/solutions/legal
8. "18 lawyers caught using AI explain why they did it", 404 Media, referenced in TechCrunch 12 May 2026 article.
9. "California issues first-of-its-kind fine against attorney who used ChatGPT to draft an appeal", CalMatters, September 2025. https://calmatters.org/economy/technology/2025/09/chatgpt-lawyer-fine-ai-regulation/
10. "Grassley scrutinizes federal judges' apparent AI use in drafting error-ridden rulings", US Senate Judiciary Committee, October 2025. https://www.judiciary.senate.gov/press/rep/releases/grassley-scrutinizes-federal-judges-apparent-ai-use-in-drafting-error-ridden-rulings
11. "Harvey reportedly raising at $11B valuation just months after it hit $8B", TechCrunch, February 2026. https://techcrunch.com/2026/02/09/harvey-reportedly-raising-at-11b-valuation-just-months-after-it-hit-8b/
12. "Legal AI startup Legora hits $5.6B valuation and its battle with Harvey just got hotter", TechCrunch, April 2026. https://techcrunch.com/2026/04/30/legal-ai-startup-legora-hits-5-6-valuation-and-its-battle-with-harvey-just-got-hotter/