Let’s talk

Insights for your business.

Staff Using ChatGPT at Work: An AI Use Policy for a Singapore Company

A computer chip marked with the letter A

Yes, a Singapore company should have a written AI use policy once staff start using ChatGPT-type tools, and the board should own it. The policy should say which tools are approved, what information may never be pasted in, who checks AI output before it reaches a client, and who answers for mistakes. PDPC’s generative AI guidelines encourage organisations to document these safeguards and to tell staff which personal data may be entered into AI systems.

A computer chip marked with the letter A
A written AI use policy helps a company control confidentiality and accuracy risks. Photo by Igor Omilaev on Unsplash.

Why does an SME need an AI use policy now?

Staff do not wait for a policy. A sales executive pastes a customer complaint into a chatbot to draft a reply. An accounts clerk uploads a supplier list to summarise it. A manager asks a tool to draft a termination letter. Each of these can disclose personal data, client confidential information or the company’s own trade secrets to a third-party service.

The Personal Data Protection Commission (PDPC) issued its Advisory Guidelines on Use of Personal Data in Generative AI on 20 July 2026. They say an organisation that adopts a generative AI system as a service, including through SaaS platforms, still bears primary responsibility for ensuring it can meet its obligations under the Personal Data Protection Act 2012 (PDPA). Buying the tool does not move the responsibility to the vendor.

For directors, the practical point is simple. If personal data leaks through a staff member’s prompt, the company is the organisation answerable under the PDPA. Our article on directors’ cyber risk liability explains how exposure after a breach can reach the board.

What does Singapore guidance actually say?

There is no Singapore statute that requires an AI use policy for a private company. What exists is guidance, and it points one way.

These documents are guidance, not law, so your policy should be proportionate. The PDPA itself, your contracts and your duty of confidence are the binding parts. For a related angle on supplier terms, see our article on a software supplier wanting customer data for AI training.

What should the policy cover?

Keep it to two pages that staff will read. These are the sections we recommend.

  1. Scope and approved tools. List the tools staff may use for work and the accounts they must use (company accounts, not personal ones). Anything else needs approval.
  2. Prohibited inputs. Personal data about customers, employees or job applicants; client confidential material; passwords and access credentials; unreleased financial results; and anything covered by a confidentiality clause.
  3. Accuracy and human review. A named person must review AI output before it is sent externally or relied on. AI can produce confident errors, including invented references.
  4. Intellectual property. Staff must not paste in third-party copyrighted material or code they have no right to use, and must record when significant content was AI-assisted. Ownership of AI output and the terms of each tool are matters to check with a lawyer, as we have not assumed an answer here.
  5. Disclosure to clients and regulators. Where a client contract restricts subcontractors or tools, AI use may need disclosure or approval.
  6. Decisions about people. No hiring, firing, discipline or credit decisions made solely on AI output.
  7. Incidents. Staff must report a wrongly pasted document at once. See our article on assessing a potential personal data breach for the assessment steps.
  8. Review and ownership. Name the owner, set a review date and record training.

Which data can go into an AI tool?

A simple traffic-light table helps staff decide in seconds. Adjust it to your tools and contracts.

Information Public chatbot account Approved business tool with contract terms
Published marketing text Allowed Allowed
Internal draft with no client or personal data Allowed if you accept the tool’s terms Allowed
Customer or employee personal data Not allowed Only for a defined purpose and after the safeguards are checked
Client confidential documents Not allowed Only if the client contract and tool terms permit
Credentials, keys, unreleased results Not allowed Not allowed

An “approved business tool” is one where the company has read the supplier’s terms on data retention, use of prompts for training, security and where data is processed, and accepted them in writing. PDPC’s guidelines say a deployer should have enough information on the system’s upstream safeguards to make that assessment.

What do the PDPA exceptions mean for staff prompts?

Some owners assume that an exception to consent solves the problem. It rarely does. PDPC’s Key Concepts guidelines (revised 29 April 2026) say that even where a PDPA exception applies, the organisation must still comply with other obligations such as protecting confidential information and contractual duties. The business improvement exception, for example, requires that the purpose cannot reasonably be achieved without individually identifiable data and that use is appropriate in the circumstances. Our guide to employee data protection for HR and payroll covers the employee side.

The safest rule for an SME is to remove names, NRIC numbers and contact details before anything is pasted into a tool, and to ask whether the task needs personal data at all.

What should the board do?

The board does not need to review every prompt. It should do five things and minute them.

  1. Resolve to adopt the policy and name an executive owner.
  2. Ask management for a list of the AI tools in use, including those staff adopted on their own.
  3. Check that the company’s data protection officer, or the person fulfilling that role, has reviewed the policy against PDPC’s guidelines.
  4. Require a short report each year on incidents, new tools and training.
  5. Check client and insurance contracts for AI-related terms.

Record the decision in a board resolution so there is evidence of oversight. Our company compliance calendar is a useful place to add the annual policy review.

Example: an accounts clerk and a supplier list

Example (hypothetical): A 20-person distribution company has no AI policy. An accounts clerk uploads a spreadsheet of customer names, phone numbers and outstanding balances to a free chatbot to draft collection emails. The upload contains personal data of individual customers, collected for billing, and the clerk used a personal account whose terms the company never reviewed.

With a policy, the clerk would know that customer data is a prohibited input for public accounts. The company could still use AI for collections by approving a business tool, removing personal identifiers, and having the finance manager review each email before sending. The policy turns a one-off judgment into a rule the whole team can follow.

Roll the policy out with the rest of your people documents. Our guide to writing an employee handbook shows where a policy like this fits and how to have staff acknowledge it.

Frequently asked questions

Is an AI use policy legally required in Singapore?

We have found no Singapore law requiring a private company to adopt one. PDPC’s guidelines encourage written policies and documented safeguards, and a written policy is the practical way to show you took reasonable steps if something goes wrong.

Can we ban AI tools instead?

You can, but bans are hard to enforce when staff use personal phones and browsers. A short approved-tool list with clear limits is usually more effective than a blanket ban.

Does the policy need to be published on our website?

PDPC’s guidelines say that making such policies available, for example on a website, helps demonstrate accountability. It is encouraged rather than required.

Who is responsible if AI output is wrong?

The company is responsible to its clients and customers for what it sends, whoever or whatever drafted it. That is why the policy requires a named reviewer. Our article on AI and Singapore’s courts shows what can go wrong when AI output is not checked.

Do we need to tell clients we use AI?

Check your client contracts and engagement terms. Where they restrict subcontractors, tools or data transfers, or require consent before client information is shared with third parties, you may need approval first.

Related reading: Using Images on Your Business Website in Singapore Without Infringing Copyright.

Need help with an AI use policy and board oversight?

If your company wants to put an AI use policy in place, contact the Raffles Corporate Services team at [email protected] or on WhatsApp at +65 8501 7133. We can help the board prepare the resolution, keep the policy in the company’s records and add the annual review to the compliance calendar. Drafting of legal terms in client or supplier contracts is a matter for a lawyer.

Willie Tan, Raffles Corporate Services

Submit a Comment

Your email address will not be published. Required fields are marked *

Real people. Right here in Singapore.

Let’s get to work.

Hop on Raffles Corporate Services