Yes, a Singapore law firm can run AI over client documents without breaching its duties, and a private system is one of the cleaner ways to do it. The Ministry of Law's guide for the legal sector (March 2026) says the confidentiality rule in the Professional Conduct Rules "does not prohibit the use of GenAI provided that appropriate safeguards are implemented". A system on your own hardware answers the hardest of those safeguard questions: whether a vendor keeps, trains on or processes client data outside Singapore. It doesn't answer the other one. Every output still has to be checked by a lawyer, and the courts have made clear who carries the blame if it isn't. Accounting firms face the same split under ACRA's new audit guidance on AI.
This post covers what a professional-services firm needs on top of a normal document AI setup. The general mechanics of searching your own documents with AI are in analysing company documents locally on GB10 and turning 10,000 documents into a private AI brain.
What do Singapore's legal-sector rules say about generative AI?
Three documents set the expectations. The MinLaw "Guide for Using Generative AI in the Legal Sector" (6 March 2026) is non-binding and rests on three principles: professional ethics (lawyers "remain ultimately responsible for all work products"), confidentiality and transparency. The Supreme Court's Guide on the Use of Generative AI Tools by Court Users has applied to all Supreme Court matters since 1 October 2024. It doesn't ban AI, but it makes users "fully responsible for the content in all your Court Documents". The Law Society's advisory of 2 April 2026 warns that public AI tools risk breaching the conduct rules, "in particular rule 6", and tells lawyers not to put privileged, confidential or personal data into them.
| Expectation | Where it comes from | What a private system changes | What it doesn't change |
|---|---|---|---|
| Know whether the vendor keeps or trains on inputs | MinLaw guide ¶23(c) | No outside vendor sees the data | Your own retention and deletion rules still apply |
| Know whether client data is processed outside Singapore | MinLaw guide ¶23(e)(iii) | Data stays on your premises | Client instructions on data handling still apply |
| Encryption, audit logs, SSO and MFA | MinLaw guide ¶23(e)(v) | You control the settings | Your firm now has to run and patch them |
| Check every output and citation | MinLaw ¶16, ¶20(c); Supreme Court guide 5(3) | Nothing | "Hallucination cannot be completely eliminated" |
| Respect confidentiality orders and court-ordered documents | Supreme Court guide 5(6), 5(7) | No external disclosure risk | Internal access still has to be restricted |
| Consider telling clients about substantial AI use | MinLaw guide ¶27, ¶30 | Easier to explain where data goes | The disclosure decision is still yours |
| Redact properly before sharing | Law Society advisory, 2 Apr 2026 | Nothing | "Blackout" tools can leave text readable |
Source: MinLaw Guide for Using Generative AI in the Legal Sector (6 Mar 2026); Supreme Court Registrar's Circular No. 1 of 2024 (effective 1 Oct 2024); Law Society advisory on publicly available AI tools (2 Apr 2026). Summaries compiled 9 Nov 2026; check the current documents before relying on them.
Why does a law firm's AI need ethical walls?
A general company assistant can often let everyone search everything. A law firm can't. Partners on opposite sides of a conflict, matters under confidentiality orders and documents produced under a court order all need to stay separated. Those court-ordered documents may be used only for the proceedings they were produced for. If your AI system indexes the whole document store and answers any question from anyone, it can quietly undo the walls your document management system enforces.
The practical rule is that retrieval must respect matter-level permissions before the model sees anything. When a fee earner asks a question, the system should only search documents that person is already allowed to open, and it should log what it retrieved. No Singapore guidance spells this out in those words. It follows directly from the confidentiality duty and the Supreme Court's rules on court-ordered material, and it is the main design difference between a law-firm system and a generic one.
A private system answers "where does our client data go?" It doesn't answer "is this output right?" That question stays with the lawyer or auditor who signs off.
How do you stop hallucinated citations reaching a court?
You can reduce them, but not remove them. The MinLaw guide names grounding and retrieval as mitigations, and says lawyers "must ensure that the output is correct" and that cited cases "remain good law". The Supreme Court guide warns that chatbots may "invent cases and statutes" and adds a rule worth printing on the wall: "you cannot use one Generative AI tool to confirm the content generated from another". Sanctions can include costs, possibly personal ones, and disciplinary action.
A private system helps by answering from your own documents and showing which passage each answer came from, so the reviewer checks a source rather than a claim. Legal research is a different job. Since September 2025 LawNet has offered GPT-Legal, a question-and-answer tool from the Singapore Academy of Law and IMDA that is reported to give referenced answers over Singapore judgments and legislation. Many firms will use a research tool like that alongside a private system for their own files.
What should go into the engagement letter?
The MinLaw guide suggests firms consider telling clients when AI use is substantial, affects costs or conflicts with the client's data or residency preferences. Engagement letters are one of the suggested channels, and Annex C has a sample clause. It also suggests offering an opt-out, noting that opting out may mean "slower turnaround time, and potentially higher costs". A firm that runs AI on its own hardware can describe its data handling in one plain sentence, which makes that conversation easier.
What about accounting and audit firms?
ACRA's Audit Practice Guidance No. 1 of 2026, "Using Artificial Intelligence Responsibly" (2 October 2026), is guidance rather than a mandatory standard, but its direction is clear. "Auditors should use only AI tools approved under the firm's governance." Where there is uncertainty, they "should avoid entering confidential or client-identifiable information". The guidance notes that different tools "may present different confidentiality, data residency and security considerations".
It names document extraction and matching records to supporting documents as process tasks AI can help with, which is where most accounting firms start: pulling figures from invoices, statements and contracts, then reconciling them. For judgement tasks, outputs must be verified "against underlying sources", and audit documentation should show how AI-assisted outputs were reviewed, including prompts, assumptions and revisions. The audit opinion can't be delegated to AI. A private system makes the confidentiality question simpler. The review and documentation duties stay the same.
What hardware does a 10–50 fee-earner firm need?
Most firms need four pieces. The first is a scanner, because scanned PDFs need text recognition before any AI can search them. The second is a compact AI system to run the model and search index. A GB10 system with 128 GB of memory suits a pilot or a small practice. The third is a NAS for the document repository. The fourth is a separate backup appliance, because an AI index is no substitute for proper records retention. Larger firms, or firms expecting many people to use the system at once, should look at a DGX Station GB300 or a GPU server. The best size depends on how many documents you index and how many users ask questions at the same time, so we size it against your matter volumes rather than a rule of thumb.
A secured cloud is also a legitimate route. MinLaw's Legal Technology Platform paired practice-management software with Microsoft 365 Copilot for Singapore law firms. A case study in the MinLaw guide describes a firm running its own on-premises model. Either can work. The choice depends on your clients' expectations and how much control your firm wants over where its files are processed.
This article is general information, not legal, medical or compliance advice. Check the current guidance and speak to your own counsel or compliance lead about your firm's obligations.
Buying in Singapore
For document capture, the Ricoh fi-8170 (S$1,847.01) and Ricoh fi-7300NX network scanner (S$1,701.55) handle daily A4 work, and the Ricoh fi-7600 (S$7,129.65) suits large bundles. For the AI system, the ASUS Ascent GX10 64GB (S$6,670), NVIDIA DGX Spark 4TB (S$9,300) and HP ZGX Nano G1n 4TB (S$11,499) cover pilots and small practices. For storage and retention, there's the Synology DS1825+ (S$2,148) and the Synology ActiveProtect DP340 backup appliance (S$6,526.80). Accounting firms may also want Dynamics 365 Business Central Essentials (S$1,095 for a 12-month licence) (listed 9 Nov 2026; prices change; ask for a current quotation). For firm-wide systems, the ASUS ExpertCenter Pro ET900N G3 and HP ZGX Fury G1n DGX Station GB300 systems are price on request. Ask us about document management or OCR software, which we don't list. To start, use Request for Quotation on the product page, call (65) 6978 3502, email Sales@sourceit.com.sg or use our contact page.
Frequently asked questions
Can Singapore lawyers use generative AI on client matters? Yes, with safeguards. The MinLaw guide says the confidentiality rule does not prohibit generative AI if appropriate safeguards are in place, and the Supreme Court allows AI use in court documents as long as users take full responsibility for the content.
Does a private AI system remove the need to check outputs? No. It keeps client data in-house, but MinLaw and the Supreme Court both require lawyers to verify outputs and citations, and the Supreme Court says one AI tool cannot be used to confirm another's output.
Do I have to tell clients that my firm uses AI? The MinLaw guide suggests considering disclosure where AI use is substantial, affects costs or conflicts with a client's data preferences, for example through the engagement letter. It is guidance, so check with your own counsel.
What does ACRA expect from auditors using AI? ACRA's October 2026 practice guidance says auditors should use only firm-approved AI tools, avoid entering confidential information when unsure, verify outputs against sources and document how AI-assisted work was reviewed.
Is a cloud AI service acceptable for a law firm? It can be. Singapore's guidance does not require on-premises AI. It asks firms to check vendor retention, training use, data location and security, which is why many firms prefer enterprise tools or their own systems for client data.
SourceIT Approved
Researched and verified by the SourceIT team for Singapore buyers.
