Law firms and notary offices work with a large volume of documents. Deeds, agreements, powers of attorney, copies of court decisions and client correspondence pile up every week. Many staff have already tried ChatGPT or Claude to summarise files or prepare drafts, often without any clear rules from the office.
AI for law firms and notaries can save real time, but two things are not negotiable: client confidentiality and accuracy. This article covers the work AI is suited to, the risks to manage, and how to control AI access down to the folder level.
Where AI helps law firms and notaries
Supporting document review
AI can help read long agreements and flag the parts that need attention, such as indemnity clauses, terms and duration, dispute resolution or one-sided obligations. The output is a list of points for an advocate or notary to check. The legal judgement is still made by the person authorised to make it.
Summarising files
Summaries are useful for thick files, for example a case chronology built from a stack of correspondence, a summary of a court decision, or the key points of a due diligence report. Ask the AI to name the source page or article for each point so it is easy to check.
Drafting
For powers of attorney, demand letters (somasi) or first drafts of agreements, AI can produce an outline based on the office's templates and the case details. That draft is only a starting point. An advocate or notary still reviews every sentence before the document is used or sent to a client.
Administrative work
In a notary office, a lot of time goes into checking whether files are complete. AI can help list the documents a client still needs to provide for a company incorporation or a title transfer, then prepare a draft message requesting them. Staff only need to check it and send it.
Some work should not be handed to AI at all, such as giving a final legal opinion, deciding case strategy or verifying the identity of the parties. That work calls for professional judgement and professional responsibility.
Client confidentiality
Client documents contain information protected by professional confidentiality duties. They also often include personal data such as NIK (national ID numbers), NPWP (tax IDs), addresses and financial details. When staff paste documents like these into a personal AI account, the office loses control over where that data goes and how long it is kept.
A few basic rules the office should put in place:
- Decide which AI services may be used, and ban personal accounts for client documents.
- Review the data retention and usage policies of the AI service you choose.
- Mask NIK, NPWP and other personal data before documents are processed by AI, in line with the principles of Indonesia's Personal Data Protection Law (UU PDP).
- Make sure there is a record of who accessed which document through AI.
- Write down the office's AI policy and share it with all staff, including interns.
- Explain to clients how AI is used in the office if they ask.
Citations and accuracy
AI models can write convincingly about articles of law that do not exist or court decisions that were never issued. For a law firm, a mistake like that is a serious risk to the client and to the firm's reputation. Every AI answer about the content of a document should therefore come with a citation pointing to its source, ideally down to the clause or page.
Get the team used to opening those citations, especially before anything goes into a legal opinion, a deed or a letter to a client. If the AI finds no basis in the document, it should say so plainly. For references to legislation and case law, always check the official source.
Keep in mind that AI reads text, and text from scanned documents is not always perfect. Numbers, dates and names in older scanned deeds should be matched against the physical document.
Controlling AI access per folder
Law firms and notaries usually store documents per client or per matter. That structure can serve directly as the basis for AI access. The office does not need to open its entire storage to AI. It only needs to decide which folders the AI may read, for whom, and for what purpose.
Say a notary office in Yogyakarta is handling company incorporations for several clients at once. A staff member working on one client opens only that client's folder to the AI. Other clients' documents are never read, and every access is logged. If a client later asks how their documents were handled, the office has a record it can show.
Set access by role as well. A partner can open the folders for every matter they handle, an associate only the matters assigned to them, and administrative staff only the admin folders. When a matter closes or someone moves team, AI access to those folders is removed too.
Start with one type of work, such as summarising case files, with a few staff and clearly defined folders. Decide who must check the AI's output, then review the access log and the quality of results after a few weeks before expanding.
Setting up AI in your office
Folders is designed for law firms, notaries, tax consultants and accountants whose staff already use AI on client documents. The office can choose which folders Claude or ChatGPT may access through MCP, with an audit log, get answers with citations down to the clause level, and mask NIK and NPWP before anything reaches the AI. You can find the details on the Folders page.
