Claude for Legal: The Complete Guide for ANZ Law Firms

Claude for Legal launched on 12 May 2026. If you are a partner or COO at an ANZ law firm, you have probably already received questions from your management committee about what it means for your practice. This guide gives you the unvarnished answer.

What is Claude for Legal?

Claude for Legal is Anthropic's legal tuned model configuration, released publicly on 12 May 2026 (Source: Anthropic product announcement, 12 May 2026). It is not a standalone product you buy off the shelf. It is a model configuration and API layer that needs to be deployed, connected to your document management system, and governed before it does anything useful for a law firm.

The distinction matters. Claude for Legal gives you a model trained with legal context and safety tuning appropriate for a professional services environment. What it does not give you is an iManage connector, a privilege classification layer, a matter management integration, or the workflow design that makes the model useful for your specific practice areas. Those are deployment decisions, and they are where most ANZ firms are currently stuck.

Anthropic positions Claude for Legal alongside Harvey and Thomson Reuters CoCounsel as a direct competitor in the legal AI market. The model handles long context documents up to 200,000 tokens, which is relevant for large discovery sets and contract portfolios. It supports structured output, meaning it can produce issue registers, clause summaries, and risk matrices rather than just free text responses.

How does it compare with Harvey and Thomson Reuters CoCounsel?

The three products operate at different layers of the market and they are not interchangeable.

Harvey is a purpose built legal AI platform built on GPT 4 with significant training data from global partner firms, including Allen and Overy. It is primarily targeted at Am Law 100 and Global 100 firms with a platform licence model and significant onboarding commitment. Harvey has deep integrations with major DMS platforms and a growing library of practice area specific agents. For most ANZ mid market firms, Harvey's pricing and implementation requirements are not matched to the scale of the opportunity.

Thomson Reuters CoCounsel sits inside the Westlaw and Practical Law ecosystem. If your firm already runs Westlaw subscriptions, CoCounsel is the path of least resistance for research and brief drafting tasks. Its strength is access to primary source materials. Its limitation is that it does not extend naturally to document management workflows or matter management integrations outside the TR ecosystem.

Claude for Legal, deployed through a structured activation, sits between the two. It does not carry Harvey's implementation overhead or CoCounsel's subscription dependency. For ANZ firms with existing iManage or NetDocuments environments and a defined use case, a Claude activation is faster to production and scoped to the actual problem.

Can Claude read documents in iManage and NetDocuments?

Yes, with a connector. Claude for Legal does not natively reach into your document management system. A properly structured activation builds an API connector between your DMS and the Claude deployment so documents are retrieved in matter context, processed, and outputs are written back to the relevant workspace or matter file.

For iManage Work 10, the connector uses the iManage REST API. For NetDocuments, it uses the ndOfficeWeb and REST API layer. Both are established integration patterns that David and Goliath has built as part of the Legal activation module.

The practical implication is that lawyers do not need to export or copy documents to interact with the AI. The agent pulls the relevant document set based on matter number, document type, or date range, processes them, and returns structured output. This is relevant for privilege management: documents that should not leave the DMS do not leave the DMS.

How does it handle privilege and confidentiality?

Claude for Legal operates within Anthropic's enterprise data handling framework, which means client matter content is not used to train future models (Source: Anthropic Enterprise Terms, May 2026). For ANZ firms with Australian Privacy Act obligations and Law Society professional conduct rules, this is the baseline, not the ceiling.

A structured deployment adds three additional layers. First, a privilege classification step runs before any document is submitted to the model, giving the supervising lawyer a decision point on sensitive materials. Second, data residency is configured to Australian AWS endpoints (Sydney region) so matter content does not transit offshore. Third, an audit trail captures every document submitted to the model and every output generated, so the supervising partner can demonstrate oversight in the event of a client or regulatory inquiry.

Ethical wall configuration is addressed during the governance module. The agent does not surface matter content across ethical wall boundaries, and for firms using IntApp Conflicts or equivalent systems, conflict status is checked as a pre condition before document retrieval is triggered.

What does a Claude for Legal deployment look like in an ANZ firm?

Most ANZ firms arrive at activation from one of two starting points. Either a senior associate has been using Claude.ai in a personal capacity and the firm wants to formalise it before something goes wrong, or the managing partner has read a competitor announcement and wants a credible response for the partnership.

A structured deployment starts with scope definition, not technology. Which practice groups generate the highest document review overhead? Where is the firm losing the most time on low complexity research that does not reach a client billable threshold? Those answers determine which use case is built first.

The first production agent is typically a document review agent for one practice area: M and A due diligence, property contract review, or employment matter document triage are the most common starting points in ANZ. The agent is connected to the DMS, given a defined issue checklist, and tested on historical matter documents before going near live client work. A supervising lawyer reviews every agent output before it is used in a client context during the initial production period.

How long until our first agent is live?

The standard activation timeline is 10 business days from the kickoff meeting to a production agent handling real documents. The timeline breaks into three phases.

Days 1 to 3 cover knowledge foundation: indexing the firm's precedent library, configuring system access to the DMS, and mapping the specific workflow the first agent will automate. Days 4 to 7 cover the agent build and integration testing. Days 8 to 10 cover testing with historical documents, lawyer review of outputs, governance documentation, and go live sign off.

Firms with non standard DMS configurations or bespoke practice management systems occasionally need an additional week for connector setup. This is identified at scope definition, not discovered mid sprint.

What does activation cost?

David and Goliath does not publish fixed prices for legal activation because the right scope depends on the firm's DMS environment, the practice areas in scope, and the number of agents being built in the sprint. The starting reference point for a single agent activation for a firm of 20 to 150 lawyers is available on the activation page.

What is consistent across all activations: the deliverable is a production agent, not a report. The sprint ends with something running in your environment, not a recommendation document about what you should build next.

What should ANZ partners do this quarter?

ANZ firms are 18 months behind global peers on legal AI adoption. That gap is closing, but not because ANZ firms are moving faster. It is closing because global firms are raising the standard of what "using AI" means, from occasional Claude.ai queries to structured agents running inside client matters.

The practical action for Q2 2026 is to identify one document heavy workflow that a senior fee earner is currently doing manually, define an issue checklist for that workflow, and run a 10 day activation against it. The result is a production agent, a governance framework your privacy officer can sign off on, and a reference point for the partnership conversation about where legal AI goes next at your firm.

To scope an activation for your firm, visit davidandgoliath.ai/claude-activation/legal or book a scoping call at davidandgoliath.ai/claude-activation/start.

Sources: Anthropic product announcement, 12 May 2026. Freshfields benchmark on document review time reduction, 2025. Thomson Reuters Future of Professionals Report, 2024. Thomson Reuters State of the Legal Market Report, 2025. Anthropic Enterprise Terms, May 2026.