Claude for Legal: What Law Firms Are Actually Doing (2026)
AI for Legal

Claude for Legal: What Law Firms Are Actually Doing (2026)

Anthropic's associate general counsel says lawyers are the second-most-active professional group on Claude, behind only software developers. Here is what they are using it for.

By AITraining2U Editorial Team 2026-08-18 11 min read
Meeting room with laptops and documents — Claude for legal work in law firms

Two things happened in the first half of 2026 that changed the legal-AI conversation. In May, Anthropic launched Claude For Legal — a dedicated law-focused plug-in with MCP connectors into the tools law firms already use. In the same month, the FTI Consulting & Relativity 2026 General Counsel Report put in-house generative AI usage at 87%, nearly double the 44% from the year before. Adoption is no longer the question. The question is what firms are actually doing with it.

This is what a Malaysian legal team should understand about Claude in 2026 — the product surface, the reference customers, and the compliance and Malaysia Bar considerations before rolling it out.

1. Why lawyers, of all people, ended up on Claude

Anthropic's associate general counsel Mark Pike told Bloomberg in May 2026 that outside of software developers, lawyers are the single most active professional group on Claude. That is not an accident. Legal work is the archetypal Claude use case: large volumes of long documents, precise language, high tolerance for a draft-then-review workflow, and a professional obligation that keeps a human accountable regardless of what tool is used.

The concrete tasks stack up quickly:

  • Drafting contract clauses from a term sheet, or from an old contract that needs updating to reflect a new position.
  • Reviewing an incoming contract against a house style guide and flagging deviations, with the specific clause and reasoning surfaced.
  • Summarising deposition transcripts, court filings, or discovery documents at scale.
  • Explaining a regulatory position in plain language for a non-lawyer stakeholder, without softening the substance.
  • Cross-referencing an internal policy document against a new regulation to identify what needs to change.

2. Claude For Legal, and how it fits with Harvey

Claude For Legal is the general-purpose product surface for lawyers — it ships MCP connectors into Harvey, Relativity, iManage, NetDocuments, and other major legal-tech platforms. That means a lawyer working inside Harvey can call Claude directly for reasoning tasks that Harvey's own models are not tuned for, and a Claude session can pull matter documents from iManage or NetDocuments without leaving the chat.

Harvey itself raised a $200 million round in March 2026 at an $11 billion valuation and now focuses on agentic legal workflow automation — bulk contract review across a deal, structured due-diligence extraction, matter management. Claude and Harvey are complementary, not competitors: Harvey is a workflow tool, Claude is a reasoning tool, and both live inside the same firm technology stack.

3. How law firms adopt Claude — the maturity ladder

Law-firm Claude adoption ladder

Law-firm Claude adoption ladder 1Stage 1Personal productivity
Individual lawyers use Claude for drafting, research, summarising. No firm policy yet. Consumer accounts on personal devices — the compliance risk to fix first.
2Stage 2Firm-approved use
Firm buys Business/Enterprise seats. Publishes a 2-page AI-use policy. Documented data-flow. Client engagement letters updated. Practice areas cleared for use.
3Stage 3Matter-specific tools
Practice-area prompt libraries. MCP into iManage / NetDocuments / Harvey. Contract-review workflows with human sign-off gates. Weekly ops review of what worked.
4Stage 4Client-facing agents
Portals where clients interact with firm-branded Claude agents for narrow, low-risk tasks (bulk NDA turnaround, contract-status queries). Full audit trail, human escalation path.

Across large firms in 2026, adoption tends to follow a predictable four-stage pattern. Firms that jump stages usually get burned; firms that walk through the stages in sequence tend to compound.

4. The confidentiality and privilege problem

The single biggest concern any general counsel raises about Claude is client confidentiality and legal professional privilege. The answer breaks into three parts:

  • Plan choice matters. Consumer Claude (Free and Pro individual) may process inputs for model improvement unless privacy settings are toggled. Claude for Business and Enterprise do not train on your data. For any privileged work, only Business or Enterprise plans are appropriate.
  • Data-flow documentation matters. Every firm should know: which matters can be discussed with Claude; which cannot; where the data is processed; how long it is retained; and who at Anthropic can access it. The Data Processing Addendum in the Business/Enterprise contract answers most of these questions.
  • Privilege is a rule of law, not a technical setting. Talking to Claude about a live matter is not a waiver of privilege in most jurisdictions, but the analysis is not simple. A general counsel should get a written opinion for the jurisdictions the firm practises in — this is a table-stakes governance item in 2026.

5. Malaysia Bar considerations

The Malaysia Bar Council has not (as of publication) issued a comprehensive AI-use practice direction, but Ruling 14.01 on client confidentiality and section 126 of the Evidence Act 1950 on legal professional privilege still apply. Practical implications for a Malaysian firm:

  • Firm-issued Claude for Business or Enterprise accounts — not consumer or personal accounts — are the minimum bar for any client-file work.
  • Client engagement letters should disclose that AI-assisted tools may be used in the delivery of legal services. This is analogous to disclosing use of paralegal support or external legal research services.
  • Every AI-drafted document that goes to a client, court, or counterparty must be reviewed and signed off by a lawyer — the responsibility does not transfer to Anthropic.
  • Cross-border data flow is worth checking. Anthropic's primary processing is US-based; firms serving clients under PDPA-sensitive sectors (financial services, healthcare, government) should confirm the transfer basis is documented.

6. What Claude does not replace

Claude does not replace the lawyer's judgement, the client conversation, the professional obligation, or the courtroom appearance. It also does not:

  • Verify facts it has not been given. If you ask Claude a question of fact and do not provide the source documents, it will draw on training data — which for legal purposes is unreliable and out-of-date.
  • Guarantee case-law accuracy. Every citation to a Malaysian, English, or other case should be verified against the actual judgment, not accepted from a Claude summary.
  • Practise law. Claude drafts, extracts, summarises, and reasons. A lawyer decides.

Firms that write these limits into their AI use policy have fewer incidents than firms that leave them implicit.

7. Starting position for a mid-sized Malaysian firm

The pragmatic starting position for a 30-100 lawyer firm in Malaysia in 2026: buy a Claude for Business subscription for the whole firm, publish a two-page AI use policy that mirrors the confidentiality and privilege rules already in the firm's engagement letters, run a pilot with the partners most inclined to try it (usually the M&A and commercial teams first), and formalise the practice areas that graduate to matter-specific Claude use over the following 60 days.

The firms that hesitate risk being the ones whose associates are already using consumer Claude accounts on personal devices to summarise judgments — without the audit trail, without the data processing agreement, and without the privilege protection the firm could otherwise arrange. That is a compliance risk being carried today, whether or not it is written down.

Training-wise, our Claude training programme for Malaysian professionals and the broader AI Automation curriculum both cover the drafting, review, and firm-policy workflows discussed here. Both are HRDC SBL-KHAS claimable for eligible Malaysian employers.

💡
HRDC SBL-KHAS claimable

Structured AI training for legal teams — including firm-policy design, prompt engineering for legal drafting, and Claude configuration — is claimable under HRD Corp SBL-KHAS for eligible employers. See how to claim HRDC funding.

About the author

AITraining2U Editorial Team →

HRDC-Certified · Practitioner-Led · Malaysia & SEA

The AITraining2U Editorial Team is a working group of practitioners — instructors, working consultants, and HRDC-certified trainers — who collectively deliver AI training to Malaysian organisations across financial services, technology, professional services, and the public sector. Articles attributed to the Editorial Team draw on consolidated learnings from live programmes, corporate engagements, and regional industry research.

Frequently Asked Questions

There is no explicit prohibition and, as of publication, no Malaysia Bar Council practice direction on AI use. The existing rules on client confidentiality (Ruling 14.01) and legal professional privilege (section 126 Evidence Act 1950) continue to apply. In practice this means a firm-issued Business or Enterprise Claude plan, a two-page AI use policy in the firm's operations manual, engagement-letter disclosure that AI tools may be used, and lawyer review of every AI-drafted document before it leaves the firm.

Both, most large firms have concluded. Harvey is a workflow product for bulk contract review, due-diligence extraction, and matter-based agents. Claude is a general-purpose reasoning tool. Claude For Legal ships MCP connectors into Harvey, so a Claude session can call Harvey for structured extraction and vice versa. The choice is not binary — it is how you compose them inside your existing legal-tech stack.

Under most common-law analyses, disclosing information to a legal-tech vendor under a contract of confidentiality does not waive privilege — analogous to using an external transcription service or legal research provider. Business and Enterprise plans include a Data Processing Addendum that supports this position. However, the analysis is jurisdiction-specific; a firm should obtain a written opinion for every jurisdiction it practises in before deploying broadly.

The heavy-document practices see the biggest lift: M&A due diligence, commercial contract drafting and review, financial services regulatory work, employment (policy drafting, dispute correspondence), and litigation document review. Estate planning and family law see moderate lift. Bar-appearance-heavy practices see the least direct lift, though drafting written submissions still benefits. Adoption typically starts with M&A and commercial, then expands.

Yes. AITraining2U's Claude and AI Automation programmes are registered with HRD Corp and claimable under SBL-KHAS for eligible Malaysian employers. Law-firm engagements are typically structured as closed-cohort in-house sessions with confidentiality provisions in the training agreement.

Want to apply this in your organisation?

AITraining2U runs HRDC-claimable corporate AI training for Malaysian organisations — from leadership awareness to hands-on builder workshops. Talk to us about a programme tailored to your team.