MinLaw's Generative AI Guide: What Singapore Law Firms Actually Have to Do
Singapore's Guide for Using Generative AI in the Legal Sector is six months old. Here's what it actually requires, and where firms are still misreading it.
The MinLaw GenAI Guide: What It Actually Requires, And Where Firms Are Still Getting It Wrong
On 6 March 2026, Singapore’s Ministry of Law published the Guide for Using Generative AI in the Legal Sector, following a public consultation that drew input from more than twenty stakeholders: law firms, in-house teams, legal tech vendors, academics and industry bodies. Six months on, it’s worth being precise about what the Guide actually says, because a fair amount of what’s circulating among Singapore firms is a paraphrase of a paraphrase.
Start with what it isn’t. The Guide is not subsidiary legislation, and it does not create new professional conduct rules. It’s guidance, and MinLaw has been explicit that it will keep updating it as technology and practice evolve. That matters practically: there is no single clause you can point to and say “we comply with section 4.” Compliance here is a posture, not a checkbox, and the firms getting the most value from the Guide are the ones that have translated its three principles into internal policy rather than treating it as a document to file away.
The Three Principles, and What They Mean in Practice
Professional Ethics
The Guide is unambiguous that a lawyer’s professional obligations don’t dilute because a machine did part of the work. You retain full responsibility for the accuracy, judgment and quality of anything that leaves your desk, regardless of whether GenAI drafted the first pass. In practice, this is the same standard that already applied to work produced by a junior associate or an outsourced provider. The Guide doesn’t invent a new duty; it confirms an old one applies to a new tool. Firms that are still treating AI output as “pre-checked” because it came from an expensive enterprise tool are misreading this principle, not failing to find it.
Confidentiality
This is the principle firms most often get wrong, because it’s usually reduced to “don’t paste client information into ChatGPT.” The Guide’s actual ask is broader: understand how the specific AI system you’re using processes data, whether inputs are retained, whether they’re used for model training, where they’re hosted, and what your vendor’s contractual commitments actually say versus what their marketing implies. A firm using an enterprise deployment with a signed data processing agreement and a firm using the consumer version of the same underlying model are not in the same position, even though the interface looks identical. If your firm hasn’t asked its AI vendors these questions in writing, “confidentiality” is currently aspirational rather than actual.
Transparency
This is the principle with the least bright-line clarity, and MinLaw appears to have left it that way deliberately. The Guide suggests disclosure to clients “especially when GenAI use substantially affects work products, costs, or involves data practices clients may question,” which is a judgment call, not a rule. The practical question firms should be asking isn’t “do we have to tell clients we use AI” but “would this particular client be surprised, or feel misled, if they found out later.” For high-stakes advisory work or novel data handling, that argues for a line in the engagement letter. For using AI to speed up a first-pass document review, most firms reasonably conclude disclosure isn’t required. The mistake is not making the judgment call at all.
What the Case Studies Are Actually For
The Guide comes with real-world case studies, and they’re worth reading properly rather than skimming. They’re the closest thing to worked examples MinLaw has given the sector, covering legal research, drafting and knowledge management deployments. If your firm is building an internal AI governance policy, borrowing the structure of these case studies (context, tool, safeguard, outcome) is more useful than starting from a blank page.
Where This Leaves Firms Now
The honest answer is that the Guide shifts the burden onto firms to document their own reasoning, not onto MinLaw to police it. That’s consistent with how Singapore has approached most professional regulation of AI so far: proportionate, principles based, and reliant on the profession’s existing culture of professional judgment rather than prescriptive rules. For firms, the practical takeaway is the same one we’ve made in the module on AI governance policy: if you don’t have anything in writing about how your firm handles the confidentiality and transparency questions above, you don’t have a policy, you have a set of ad hoc decisions that will be very hard to defend consistently if a client, or MinLaw, ever asks.
The tool is not the advantage. The firm that has actually worked through what “understanding how the AI system processes information” means for its specific stack is.