Last updated 24 October 2026 — by Adrian Tan, SDM. Marketing guidance, not legal advice. References are to the Consumer Protection (Fair Trading) Act 2003 and the Securities and Futures Act 2001 as shown on Singapore Statutes Online on 24 October 2026, to the Singapore Code of Advertising Practice as published by ASAS, and to ASAS, CCS and MAS publications up to that date. Where we interpret how a rule applies, we say so.
A CRM vendor’s homepage says “AI-powered lead scoring”. A skincare brand’s Instagram promises an “AI skin analysis” that recommends the perfect serum. A tuition app claims its “AI tutor” gets students “two grades higher”. A property portal advertises “AI valuations”. A robo-adviser markets “machine-learning portfolios that beat the market”.
Some of these claims are accurate. Some describe a rules-based filter or a spreadsheet formula with a new label. Overstating what AI does in a product has a name, AI washing, and in the United States regulators have already fined companies for it. Singapore has no law aimed at AI claims, and as at October 2026 we found no published Singapore ruling against an “AI-powered” claim. That does not leave the claim unregulated. “AI-powered” is a statement about how a product works, and Singapore’s consumer-protection law and advertising code already cover that kind of statement in full.
This guide is for marketers, founders and agencies writing copy about AI features. It covers the legal basis, what ASAS and CCS have said so far, comparative and results claims, AI-generated testimonials, financial services, what proof to keep on file, and four worked examples.
The short version
- “AI-powered” is a factual claim. Under the CPFTA, saying a product has characteristics or benefits it does not have is an unfair practice.
- You need the proof before the ad runs. The advertising code requires claims about objective facts to be capable of substantiation, with the evidence ready to produce without delay.
- Technical jargon does not prove anything. The code bars using scientific terms to give a claim a basis it does not have.
- A disclosure label is not a defence. ASAS has said misleading content is unacceptable regardless of the tools used or the disclosures added.
- Results and “smarter than” claims need the most evidence. Comparisons must be fair and like-for-like; testimonials must be genuine and typical.
- Financial products carry criminal exposure. A false statement likely to induce someone to buy securities or fund units can breach section 199 of the Securities and Futures Act.
What “AI washing” means
AI washing is the AI-era version of greenwashing: describing a product as using artificial intelligence, or as being more capable because of AI, when that is untrue or exaggerated. It takes several forms:
- Labelling ordinary automation as AI. A fixed if-then rule, a keyword filter or a scheduled email sequence described as “AI”.
- Overstating what the AI does. A product uses a third-party language model to draft one email field, and the ad says the whole platform “thinks for you”.
- Implying autonomy that is really human. “AI-generated reports” that are written by staff, or “instant AI answers” that are routed to a call centre.
- Unproven results. “Our AI doubles your sales” or “95% accurate” with no test behind the number.
- Implied endorsement. Suggesting a regulator or government programme has approved the AI.
The common thread is that a buyer is paying, at least in part, for a capability that is not there. That is why the existing rules on misleading claims apply directly.
The CPFTA: “AI-powered” is a performance claim
Section 4 of the Consumer Protection (Fair Trading) Act makes it an unfair practice for a supplier, in a consumer transaction, to do or say anything (or omit anything) that might reasonably deceive or mislead a consumer, to make a false claim, or to do anything listed in the Second Schedule. Four items on that list fit AI claims closely.
| Second Schedule item | What it covers | AI example |
|---|---|---|
| Para 1 | Representing that goods or services have performance characteristics, components, qualities, uses or benefits they do not have | “AI-powered fraud detection” on a tool that uses fixed rules |
| Para 4 | Representing that goods are of a particular standard, quality, grade or method of manufacture when they are not | “Designed by AI” for a product designed conventionally |
| Para 16 | Presenting an ad in an objective form, such as an editorial or scientific report, without saying it is an ad | A sponsored “independent AI benchmark” article that is really marketing |
| Para 23 | Omitting or concealing a material fact, including by using small print | “AI answers in seconds” when most queries go to human staff, disclosed only in the terms |
The CPFTA bites only in consumer transactions, so a purely business-to-business SaaS sale sits outside it. B2B claims still fall under the advertising code and, where relevant, the general law of misrepresentation. For consumer sales, a consumer can sue the supplier, with claims up to $30,000 under section 6, and under section 9 the Commission can ask the court for a declaration or injunction against a supplier engaging in an unfair practice.
In our reading, the test for an AI claim is the same as for any characteristic: would a reasonable consumer understand the claim to mean something about the product that is not true? If your ad says “AI recommends the right size”, a consumer expects the recommendation to come from a model that uses data about them, not from a fixed size chart.
The advertising code: substantiation, jargon and presentation
The Singapore Code of Advertising Practice (SCAP), administered by the Advertising Standards Authority of Singapore (ASAS), applies to ads in all media. The edition ASAS publishes on its site is the 2008 code, and these are the clauses that matter most for AI claims.
- Part II, clause 5.1. Ads should not mislead by inaccuracy, ambiguity, exaggeration, omission or otherwise, including about the quality of the product.
- Part II, clause 5.3. Ads should not misuse research results or technical publications, should not present statistics as more valid than they are, and should not use scientific jargon to give a claim a scientific basis it lacks. This is the clause aimed at “neural”, “deep learning” and “proprietary algorithm” copy that sounds impressive but proves nothing.
- Part III, clause 1.1. Descriptions, claims and comparisons about objectively ascertainable facts must be capable of substantiation, and advertisers and agencies must hold that substantiation ready to produce without delay.
- Part III, clause 2.1. Where a claim relies on research, the researchers must accept the ad as an accurate account of it.
The practical effect of clause 1.1 is that the evidence file must exist before the ad runs. If ASAS receives a complaint, it asks the advertiser for an explanation and substantiation. “Our engineers are confident” is not substantiation; a documented description of the system and test results are.
What ASAS and CCS have said about AI so far
ASAS: the tool does not matter, the content does
In a media release on 23 March 2026, ASAS reported 379 pieces of feedback in 2025, of which 7 concerned generative AI in advertising, against 3 across 2023 and 2024 combined. It said the use of AI is not objectionable in itself and that its approach is independent of the tools used. Its central statement was that false and misleading content is unacceptable “regardless of the tools used or the disclosures added”. One example in the release was a telemedicine ad using a generative-AI testimonial video, which was withdrawn. ASAS also called on industry bodies to develop sector-specific guidelines on AI.
CCS: transparency about capabilities
The Competition and Consumer Commission of Singapore, which has used the short form CCS since 25 September 2025, has published two documents that bear on AI claims.
- The AI Markets (AIM) Toolkit. Its fact sheet, last updated 1 October 2025, sets a transparency principle calling for clear and truthful communication about AI capabilities and accurate descriptions of an AI system’s capabilities and limitations, and notes that false or misleading claims could be an unfair practice. CCS says it will not endorse businesses that use the toolkit, so “CCS-approved AI” is not a claim anyone can make.
- The Guide on Quality-Related Claims. Issued on 6 October 2025 with greenwashing in mind, it sets five principles: accuracy, clarity, relevance (including fair, substantiated comparisons), supporting information and evidence that is credible, specific and current. It does not mention AI, but the principles transfer directly. Our guide to green claims in Singapore marketing shows how they work in practice.
Where AI has already led to enforcement: fake reviews
CCS’s AI-related enforcement so far has been about AI used to fake social proof, not about AI capability claims. On 3 July 2025 it announced action against Lambency Detailing, whose operator had used ChatGPT to write fake five-star reviews on Sgcarmart over about two years. On 30 September 2026 it announced action against 45 businesses and a fake-review provider that used generative AI to imitate real customer reviews; the businesses gave undertakings and published apologies, and two that refused remained under investigation. If you use AI anywhere near reviews, read our guide to getting genuine Google reviews in Singapore first.
Comparative claims, results claims and AI testimonials
“Smarter than”, “more accurate than”, “10x faster”
SCAP Part III clauses 9.1 to 9.3 require comparisons to be clear and fair, to compare on the same basis and not to cherry-pick. An AI comparison is only fair if both products were tested on the same tasks, the same data and the same conditions, and the ad says what was compared. “Twice as accurate as Brand X’s chatbot” based on ten questions you chose is the cherry-picking the clause bars. A claim like “10x faster” needs a stated baseline: faster than what, measured how?
Results promised by AI
Claims that AI will raise grades, sales, rankings or returns are the riskiest on the ladder, because they promise an outcome that depends on many factors. SCAP clause 3.4 bars presenting exceptional results as typical and requires the advertiser to hold evidence that a testimonial reflects typical experience. If your best client doubled sales with your AI tool and the median client saw a modest lift, the ad cannot imply the doubling is normal.
AI-generated testimonials and avatars
SCAP clause 3.1 requires testimonials to be genuine and based on real personal experience, and clause 3.3 bars fictitious characters that imply they are real people. An AI avatar reading a made-up customer story fails both, however realistic it looks, and ASAS’s telemedicine example shows it will act. An AI avatar used openly as a presenter, not as a customer, is a different matter, but the claims it makes must still be true. Our guide to using AI in content creation in Singapore covers the production side.
Financial services: the higher stakes
For financial products, an exaggerated AI claim can move from a consumer-protection problem to a criminal one. Section 199 of the Securities and Futures Act prohibits making a statement that is false or misleading in a material particular and likely to induce people to subscribe for or buy securities or units in a collective investment scheme, where the person does not care whether it is true or knows or ought reasonably to know it is false. “Our AI picks winning stocks” for a fund that is in fact managed by conventional analysis is the kind of statement the section covers.
MAS’s FEAT principles, issued in 2018, ask financial institutions to disclose their use of AI and data analytics to data subjects proactively. MAS issued Guidelines on AI Risk Management on 7 October 2026, taking effect a year later, and its December 2024 information paper on AI model risk management dealt with internal governance. We found no MAS statement specifically on AI washing in marketing, so we do not claim one. The US shows where this can go: on 18 March 2024 the SEC settled charges against two investment advisers, Delphia and Global Predictions, for false AI claims, with penalties of $225,000 and $175,000. Our guide to investment offer advertising in Singapore covers the wider rules.
Disclosure is not a defence
A common assumption is that a small “AI-assisted” label or a disclaimer in the footer solves the problem. It does not. ASAS’s position is that misleading content stays unacceptable whatever disclosures are added, and CPFTA paragraph 23 treats hiding a material fact in small print as an unfair practice in itself.
Two kinds of disclosure are worth keeping separate. The first is telling people what your AI does with their data, a PDPA notification duty, explained in PDPC’s March 2024 guidelines on AI recommendation and decision systems and covered in our guide to AI chatbot rules for Singapore websites. The second is the claim your ad makes about the AI, which must be true on its own terms. Good data notices do not rescue an exaggerated ad.
How to substantiate an AI claim: what to keep on file
| Claim type | What to keep on file before the ad runs |
|---|---|
| “Uses AI” / “AI-powered feature” | A plain description of which feature uses AI, what kind (in-house model, third-party model via API, rules), and what it does that a non-AI version would not |
| Performance (“faster”, “more accurate”) | Test method, dataset, baseline, date, results, and who ran the test |
| Comparison with a competitor | Like-for-like test on the same tasks and conditions, the version of each product tested, and the test date |
| Results (“more sales”, “higher grades”) | Data across all users or a fair sample, the typical result, and the basis for any testimonial being typical |
| Third-party validation | The validator’s written report and consent to the wording (SCAP cl 2.1) |
Keep the file updated. AI products change often, and a claim that was true at launch can become untrue after a vendor swap or a model change. The same discipline applies to other technical claims; our guides to cybersecurity claims in marketing and ISO and certification claims take the same approach.
Four worked examples
1. The CRM with “AI lead scoring”
The scoring is a weighted points formula the client sets manually. Calling it “AI” risks CPFTA paragraph 1 if sold to consumers and SCAP clause 5.1 in any case. Fix: call it “automated lead scoring”, or add a genuine model and document what it does.
2. The skincare “AI skin analysis”
The app does use an image model, but the “perfect serum” recommendation is drawn from a short list of the brand’s own products. Fix: “AI skin analysis that suggests products from our range”, and no claim that the result is clinically accurate without data. Any health or efficacy claim also needs to meet the cosmetics and health-product rules.
3. The tuition app’s “two grades higher”
The claim comes from three students. Under SCAP clause 3.4 that is an exceptional result presented as typical. Fix: report the result across all users with a stated period, or drop the number and describe the feature.
4. The robo-adviser that “beats the market”
This combines a results claim with a financial product, so section 199 of the SFA and the financial-advertising rules are in play. Fix: describe what the model does, state that past performance does not guarantee future returns, and have the copy reviewed by compliance before release.
Where this leaves you
Singapore has not written an AI-advertising law because it does not need one to deal with AI washing. An AI claim is a claim about a product, and the CPFTA and the advertising code already require it to be true, clear and provable. ASAS has said the tool does not matter, CCS has asked for accurate descriptions of AI capabilities, and enforcement has already reached AI-written fake reviews. Describe what your AI actually does, test before you compare, keep the evidence file current, and the AI label will help rather than hurt you.
If you want campaigns whose claims, tracking and results hold up, our performance marketing team in Singapore can help, and you can see how we report results in our client case studies. For the wider picture, read our complete guide to performance marketing in Singapore.
Frequently asked questions
Is it illegal to call my product “AI-powered” in Singapore?
No, if it is true. It becomes a problem when the product does not use AI in the way the claim suggests. Under the CPFTA, representing that a product has characteristics or benefits it does not have is an unfair practice in consumer transactions, and the advertising code bars misleading claims in all ads.
What evidence should I keep for an “AI-driven” claim?
A written description of which feature uses AI, what type and what it does, plus test data for any performance, comparison or results claim. The advertising code requires substantiation for factual claims to be ready to produce without delay, so assemble it before the ad runs.
Does adding an “AI-generated” label make an ad compliant?
No. ASAS has said false and misleading content is unacceptable regardless of the tools used or the disclosures added. A label can help transparency, but it does not cure a misleading claim, and hiding a material fact in small print is itself an unfair practice under the CPFTA.
Can I say my AI is smarter or more accurate than a competitor’s?
Only with a fair, like-for-like comparison. The advertising code requires comparisons to be clear and fair, made on the same basis and not cherry-picked. State what was compared and keep the test data, the product versions and the test date.
Can I use AI-generated testimonials or avatars in ads?
Not as fake customers. The advertising code requires testimonials to be genuine and from real personal experience, and bars fictitious characters that imply they are real. ASAS has already had an ad with a generative-AI testimonial video withdrawn. An openly AI presenter is different, but its claims must still be true.
Has any Singapore regulator penalised a business for AI washing?
As at October 2026 we found no published ASAS, CCS or MAS case on an AI-capability claim. CCS has acted against businesses using generative AI to fake customer reviews, including 45 businesses in September 2026, and in the US the SEC and FTC have penalised false AI claims.
