- 21
- October
"What can Claude actually do for a legal department or a law firm?" — the short answer is almost every "read, compare, draft, summarize" task a lawyer does, but not the "decide" tasks, and nothing that touches client confidences without permission. This article lists ten jobs that in-house legal teams and Thai law firms can hand to Claude today, the line between data you may paste and data you must not, and the rules in the Thai Chief Justice's 2025 guidance on AI that apply to anyone who files documents with a court.
In one line: Claude for legal work = a very fast reading and drafting assistant plus a lawyer who checks every line before anything leaves the office · use a Team or Enterprise plan, which does not train on your data by default · client confidences without permission never go into AI · anything filed with a court must disclose AI use, every cited precedent must be verified, and the filer stays responsible.
Before you read: Tool facts come from the Legal plugin page on claude.com, Anthropic's claude-for-legal repository and press coverage of the 12 May 2026 launch. Court rules come from the Thai Chief Justice's Guidance on the Use of Artificial Intelligence in Case Work B.E. 2568 (2025) as announced by the Court of Justice spokesperson on 9 October 2025, and from the analysis in the Dulapaha law journal. Everything was checked on 21 October 2026. Prompt examples and time savings are examples from work with our customers, not statistics. This article is general information, not legal advice.
1. Why legal teams get started with AI faster than other departments
Legal work is almost entirely text: contracts, case files, regulations, correspondence, memos. That is exactly what language models do best, and Claude reads several hundred pages in one pass (see what a context window is). Accounting or warehouse teams first have to pull data out of a system. Legal teams already have the PDFs in hand and can start on day one.
Market numbers point the same way. The Clio Legal Trends Report 2025 found that 79% of legal professionals use AI (Artificial Intelligence) in some form, flat against 2024, rising to 87% in large firms. Thomson Reuters' Future of Professionals 2025 reports that legal professionals who use AI expect to free up about 240 hours per person per year, up from 200 hours in 2024. That is a forecast by survey respondents, not measured time, but it shows which tasks practitioners believe AI helps with.
Anthropic treats legal as a core vertical. Early in 2026 it shipped a Legal plugin for Claude Cowork with ready-made commands: /review-contract checks a contract clause by clause against a playbook with green, yellow and red flags, /triage-nda sorts incoming NDAs (Non-Disclosure Agreements), plus /vendor-check, /brief and /respond. On 12 May 2026 it expanded this into "Claude for Legal": twelve practice-area plugins and more than twenty connectors to legal tools such as Thomson Reuters CoCounsel Legal, Harvey, Box, Everlaw, DocuSign and iManage. The sentence Anthropic puts in bold in its repository is "Every output from these plugins is a draft for attorney review — not legal advice."
For Thai firms, those plugins and connectors are still built around US law and US databases. As of our check on 21 October 2026 there is no connector to a Thai case-law database. What works immediately is the model itself, your own documents, and the plugins' working structure, which translates into Thai well.
Using Claude at work? Get a quote in Thai Baht with a full tax invoice
A short form — you get a request number by email right away, and our sales team follows up using that number · one form for Team and Enterprise
2. The rules before you start: the Chief Justice's guidance and the lawyers' code of conduct
The Thai Court of Justice has not banned AI. It set a framework in the "Guidance of the President of the Supreme Court on the Use of Artificial Intelligence Technology in Case Work B.E. 2568", announced publicly by the Court of Justice spokesperson on 9 October 2025. The framework covers judges, court staff and the parties who file documents, so every law firm is inside it from the first pleading drafted with AI help. Lawyers also sit under the Lawyers Council Regulation on Lawyers' Etiquette B.E. 2529 (1986). The table turns both into working rules for Claude.
| Rule | Source | What it means when working with Claude |
|---|---|---|
| Parties must disclose to the court that AI-generated material was used to draft a document | Chief Justice's Guidance B.E. 2568 (section on parties) | Add a field to your internal checklist, "which parts were drafted with AI", and include a disclosure statement in filings as each court's practice requires |
| Always verify AI output, including the drafted text, the statutes and the Supreme Court precedents it cites | Guidance clause 4 and the warning to parties | Every case number and every section number gets checked against the original before use, no exceptions |
| The filer remains responsible for AI-generated content | Guidance | The name signed at the bottom is the lawyer's. Responsibility sits with a person, not a tool |
| Take care before putting evidence containing personal or sensitive data into a prompt, since it may be disclosed to third parties unintentionally | Warning to parties, and clause 2 for court work | Strip names, ID numbers and addresses, or use pseudonyms, before pasting case material into Claude |
| AI supports administrative and academic work. Humans exercise discretion and decide | Guidance clauses 3 and 4 | Let Claude structure, summarize and compare. Legal conclusions belong to the lawyer |
| A lawyer must not disclose a client's secrets learned in the course of duty unless the client permits or the court orders | Lawyers Council Regulation on Lawyers' Etiquette B.E. 2529, clause 11 | Confidences the client has not released never go into AI on any plan. Written client consent solves this |
| No false statements, documents or evidence, and no tricks that mislead the court | Regulation clauses 7 and 8 (as analysed in the Dulapaha journal) | A fabricated precedent that reaches the court is not the AI's fault. It is the lawyer's disciplinary liability |
Why "verify every precedent" is not an exaggeration: Damien Charlotin's AI Hallucination Cases database collects court decisions worldwide in which a court found that a party relied on AI-fabricated material, mostly precedents that do not exist. It stood at 2,202 cases on 19 October 2026: 872 filed by lawyers, 1,267 by self-represented litigants, with 1,501 in the United States, 234 in Canada and 112 in Australia. There is no Thai case in the database yet, but the Chief Justice's guidance wrote the rule in advance.
3. The data line: what may go into Claude and what must not
Plans first. Claude Team and Enterprise do not use your inputs or outputs to train models by default (privacy.claude.com, updated 18 August 2026). The one exception is the thumbs up/down rating button: when a user presses it, Anthropic may store the entire conversation for up to five years and use it for model improvement. An Owner can switch the button off for the whole organization under Organization settings > Data and Privacy, "Rate chats". A law firm should do that on day one.
Location next. Data is processed on Anthropic's systems in the United States, not on the firm's own server. Sending personal data abroad falls under section 28 of Thailand's PDPA (Personal Data Protection Act), so you need a DPA (Data Processing Agreement) with Anthropic. Where to click and what to sign is in Configure Claude Team for Thailand's PDPA, and the security overview is in Claude Data Governance & Security. If you need nothing stored at all, see Zero Data Retention, which covers API and Claude Code use, not the chat interface.
| Type of data | Allowed? | Condition |
|---|---|---|
| Statutes, regulations, published judgments | Yes | Attach the original file and instruct Claude to answer from the file, not from the model's memory |
| Contract templates, negotiation playbook, firm policies | Yes | Load them as Project knowledge and reuse across matters |
| Real counterparty or client contracts | Yes, on Team or Enterprise | Remove personal data not needed for the review, such as ID and bank account numbers, and read the contract's own confidentiality clause for restrictions on third-party processors |
| Case files, witness statements, documentary evidence | With care | Pseudonymize as the court warns. Get written client consent and keep it in the matter file |
| Sensitive data under PDPA section 26, such as criminal records, health and biometric data | No by default | Only with a lawful basis and after identifiers are removed |
| Client confidences the client has not released | No | Etiquette Regulation clause 11 has no exception for plans or tools |
What works in practice: turn the table above into a one-page data rule for the firm, pin it inside every Project, and run a 60-minute training. Most people who hesitate to use AI are not afraid of the tool. They do not know what they are allowed to paste. Once the line is clear, they start.
4. Ten use cases that work today
The table separates jobs that suit an in-house legal department from jobs that suit a law firm, and every row has a "who checks" column, because that is where the rules in section 2 do their work.
| # | Job | What Claude does | Who checks, where it ends | Best for |
|---|---|---|---|---|
| 1 | Contract review against a playbook | Reads the counterparty's contract clause by clause, compares it with your standard positions, flags green, yellow or red, proposes replacement wording and cites page numbers | A lawyer reads every yellow and red flag and spot-checks the greens | Both |
| 2 | Triage of incoming NDAs | Sorts each NDA into three piles: standard, approve · a few clauses need counsel · full review, with a reason per document | The team lead approves pile one; the rest join the lawyers' queue | In-house legal |
| 3 | Summarizing case files and evidence bundles of several hundred pages | Builds a fact timeline, a list of people and roles, and a list of documents referenced in the file but missing from the bundle, every line with a page number | The lawyer on the matter, as a map for reading the file, not a substitute for reading it | Law firms |
| 4 | Drafting standard letters | Demand letters, notices, replies to notices, answers to questions from other departments, from the firm's templates and the facts provided | The signing lawyer | Both |
| 5 | Responding to PDPA rights requests | Drafts the reply to a DSAR (Data Subject Access Request) within the 30-day deadline and a checklist of data to collect from each system (see PDPA Data Subject Access Requests 2026) | The Data Protection Officer | In-house legal |
| 6 | Checking procurement terms | Compares payment milestones, penalties, performance security and warranty clauses in purchase or service contracts against policy and the approved purchase order | Legal together with procurement | Companies and public agencies |
| 7 | Hearing preparation | Extracts the issues in dispute, drafts cross-examination questions from witness statements, groups documents by issue, and checks dates and figures for consistency across documents | The lawyer, and the filing must disclose AI use under the guidance | Law firms |
| 8 | Monitoring new laws and notifications | Weekly summary of notifications, ministerial regulations and bills that affect the business, compared with current policy to show the gaps (example topic: Thailand's draft AI law) | The compliance owner | In-house legal |
| 9 | Legal memoranda | Structures the memo in IRAC form (Issue, Rule, Application, Conclusion) from the facts and the statutory text the lawyer supplies | The lawyer supplies the law and the conclusion; Claude supplies structure and language | Both |
| 10 | Running the office | Turns client meetings into minutes with next actions, drafts invoice narratives from time entries, and pulls court dates and deadlines out of email into a schedule | The office manager | Law firms |
Rows 1, 2 and 4 are what Anthropic's Legal plugin already does with /review-contract, /triage-nda and /respond. The playbook itself, standard positions, acceptable ranges and escalation triggers, is a settings file you write. The Claude for Legal suite splits twelve plugins by practice area, such as commercial, corporate, employment, privacy, intellectual property, litigation and AI governance. All of it is designed around US law, so the working structure transfers but every legal rule has to be swapped for the Thai one. Rows 3 and 7 need no plugin at all: attach the files and instruct precisely (file-reading limits are in analysing files with Claude).
5. Three reusable prompts
All three prompts share three constraints: cite page numbers, write "not found" when something is absent from the document, and never cite law that was not attached. Those three lines remove most of the fabricated-precedent risk. Pseudonymize real documents as in section 3 before pasting.
Contract review against a playbook
You are a legal assistant at [company]. Read the attached contract clause by clause and compare it with our positions in playbook.md. Produce a table: Clause | Topic | Contract wording (page) | Our position | Flag green/yellow/red | Proposed wording Rules: - If the playbook covers a topic the contract does not, write "not found". Do not invent text. - Do not cite any law or judgment that is not in the attached files. - End with the three items a lawyer must look at before this goes back to the counterparty.
Triage of incoming NDAs
Read the attached NDA and place it in one pile: A matches our standard (criteria in nda-standard.md), can be approved B differs from standard only in acceptable clauses, counsel reviews those clauses only C needs full review State the pile, a reason in at most five lines, and the list of clauses that differ from standard with page numbers. If unsure, choose C.
Case-file timeline
From the attached case documents (names already replaced with pseudonyms), do three things: 1. Fact timeline: Date | Event | Source document (page) 2. List of people and their roles 3. Documents referenced in the file but not included in the attached set Every line must carry a page reference. Do not assess which side is right. If two documents give conflicting dates, show both values with their pages.
The basics of instructing Claude that apply to every prompt are in the prompting guide.
6. Setting Claude up safely for a legal team
| Setting | Where | Why |
|---|---|---|
| One Project per matter or contract type, with the playbook, templates and frequently used statutes as starting knowledge | Claude Projects | Answers come from your documents, and matter A's data never mixes with matter B's |
| A Skill called "contract review, our way" holding the steps and the table format | Claude Skills | Everyone on the team gets the same output shape regardless of who typed the prompt |
| Turn off Rate chats | Organization settings > Data and Privacy (Owner) | Prevents conversations being kept for five years through the rating button |
| Roles and permissions: who sees which Project, who is Owner | Team admin policy | Matter confidentiality does not leak across teams, and someone owns the settings |
| Cowork with the Legal plugin for work on local files | Claude Cowork on the desktop app | Review a folder of contracts in one batch without uploading file by file |
| Claude inside Google Docs if the firm runs on Google Workspace | Claude for Google Workspace | Draft and edit in the original file instead of copying back and forth |
| A one-page data rule, a 60-minute training, and a look at usage numbers every Friday | 30-day rollout plan | The seats you bought get used instead of sitting idle for eleven months |
7. Legal departments in organizations that run Saeree ERP
In companies and public agencies, contracts do not live only in the legal department's files. The real vendor records, purchase orders, payment milestones and performance securities sit in the ERP (Enterprise Resource Planning) system. The approach we recommend to customers today is simple: legal uses Claude to review the contract file as in rows 1 and 6, then records the agreed terms back into Saeree ERP, which stays the source of truth. Claude is the reading assistant. The ERP is where the facts live.
Saeree ERP has no AI built in. What Grand Linux does is connect Claude to the ERP over MCP (Model Context Protocol, the open standard that lets an AI call external systems) with role-based permissions and a log of every call. A legal-team example: ask "which service contracts have performance securities expiring within 60 days, and who are the counterparties?", and Claude reads the ERP within the asker's own permissions, with no request to procurement for a report. Details are at Claude Solutions. This is work for after the team is fluent with Claude and the data rule has settled. Law firms without an ERP can skip this section entirely; Projects and Cowork are enough.
8. Where it fits and where it does not yet
- Claude works well for reading long documents you already have, comparing against written standards, drafting from templates, structuring memos and summarizing meetings. Every task where the person doing it can check the result before it goes out.
- Not yet for sending answers to clients or opposing parties without a lawyer's review, letting Claude recall case numbers from its own memory, pasting client confidences without permission, pasting sensitive data before identifiers are removed, or using the output to decide a dispute in place of human judgment.
- On plans: all ten jobs touch real contracts and client data, so they belong on an organization plan, Team or Enterprise, with central administration, not on individual lawyers' personal accounts. Use cases for other departments are in 12 real use cases by department, and the plan comparison is in Team or Enterprise.
Conclusion
- Legal teams start fast with Claude because the work is text and the documents are already in hand. The ten jobs here work today without waiting for Thai plugins.
- The Chief Justice's Guidance B.E. 2568 sets three core rules for parties: disclose AI use, verify the output including cited precedents, and stay responsible for the content. Clause 11 of the Lawyers' Etiquette Regulation adds client confidentiality.
- The database of AI-fabricated citations stood at 2,202 cases worldwide on 19 October 2026, 872 of them filed by lawyers. "Verify every precedent" is not an exaggeration.
- Use Team or Enterprise, turn off Rate chats, sign the DPA, pseudonymize case material, and write a one-page data rule.
- Claude drafts, the lawyer checks, the ERP keeps the facts. Connecting the ERP over MCP comes after the team is fluent.
AI drafts faster than anyone in the office, but the name signed at the bottom of the document is still the lawyer's.
- Saeree ERP team, Grand Linux Solution
References
- Anthropic — Legal plugin for Claude Cowork (commands /review-contract, /triage-nda, /vendor-check, /brief, /respond; "All outputs should be reviewed by licensed attorneys")
- Anthropic — claude-for-legal repository: the twelve practice-area plugins, connectors, and "Every output from these plugins is a draft for attorney review"
- TechCrunch (via Yahoo Tech) — Anthropic expands Claude legal AI tools for law firms (12 May 2026)
- Damien Charlotin — AI Hallucination Cases database (2,202 cases as of 19 October 2026)
- THE STANDARD — Court of Justice issues the Chief Justice's guidance on AI in drafting pleadings: disclose and take responsibility (9 October 2025, Thai)
- Daily News — Parties warned on using AI to draft pleadings: disclose, verify, take responsibility (Thai)
- Dulapaha Journal, Vol. 72 No. 3 (Sep–Dec 2025) — Dr. Anuporn Aroonrut, The Adaptation of Thai Lawyers in the Age of Artificial Intelligence: An Analysis of the Recommendation of the President of the Supreme Court (PDF, Thai)
- Lawyers Council of Thailand — Regulation on Lawyers' Etiquette B.E. 2529 (PDF, Thai)
- Anthropic Privacy Center — Is my data used for model training? (updated 18 August 2026: commercial-plan default, the rating-button exception kept up to 5 years, and how to disable Rate chats)
- Clio — 2025 Legal Trends Report (79% of legal professionals use AI)
- Thomson Reuters Institute — Future of Professionals Report 2025 (about 240 hours per person per year expected)
Is your legal team ready to start with Claude?
Grand Linux supplies Claude Team and Enterprise to companies and law firms in Thailand, with quotes in Thai Baht and a full tax invoice. Submit a request and our sales team will contact you, referencing your request number.
Get advice / request a quoteTel 02-347-7730 | sale@grandlinux.com
