Part of the AI Regulation News hub.
Singapore's digital ministry has told Parliament in writing that the Online Safety Commission's jurisdiction over inauthentic material abuse has not commenced, and that details of when it will are still to come
The useful part is not what Singapore has done. It is a minister confirming, in a written parliamentary answer that binds nobody and changes no law, which door is shut, which is the fact a victim's adviser needs before advising anyone to knock on it.
Bottom line: This is a ministerial written answer, not an instrument. It binds nobody and changes no law. Its value is that it states the government's position on the record: relief for inauthentic material abuse under OSRAA is not yet commenced, and no commencement date is given.
Who this affects: Media and entertainment rights holders and their counsel, advertisers and agency legal teams, trust and safety and content moderation leads at online services operating in Singapore, and lawyers advising individuals whose likeness or voice has been used without consent.
Issue date: 10 September 2026. The answer was given to a question for written answer at the Parliament sitting of that date, and published on MDDI's newsroom under the same date.
What changed: Nothing in law. What changed is the state of public knowledge: the ministry confirmed which OSRAA harms the Online Safety Commission can act on today and which one it cannot.
Analysis: The five live harms are defined by what the material does to a person, not by how it was made, and the ministry says the Commission can act on AI-generated material that meets those thresholds. So the gap is narrower than no protection and wider than full coverage.
Primary sources: MDDI's Response to PQ on Protecting Against Unauthorised AI-Generated Likeness in Advertisements, 10 September 2026
- Instrument (EN)
- Ministerial written answer to Parliamentary Question No. 29 on unauthorised AI-generated likeness in advertisements and the Online Safety Commission's mandate over inauthentic material abuse
- Authority
- Ministry of Digital Development and Information, Singapore
- Jurisdiction
- Singapore
- Status
- Published answer to a question for written answer. Not an instrument, not a rule, not guidance
- Bindingness
- Non-binding. It creates no obligation and confers no right. It is a statement of the government's position and of the commencement status of an existing Act
- Issue date / next deadline
- 10 September 2026. No deadline. The ministry said it will provide more details on commencement in due course, with no date attached
- Legal basis
- Refers to the Online Safety (Relief and Accountability) Act, the Online Criminal Harms Act and the Singapore Code of Advertising Practice
- Primary source
- https://www.mddi.gov.sg/newsroom/mddi-response-to-pq-on-protecting-against-unauthorised-ai-generated-likeness-in-advertisements-and-operationalising-online-safety-commission-mandate-over-inauthentic-material-abuse/
What was asked, and what was answered
The question, from Mr Yip Hon Weng, had three limbs: whether platforms will be required to verify advertisers and obtain consent attestations before publication, whether persistent disclosure and expedited removal will be mandated, and when the Online Safety Commission's powers over inauthentic material abuse will commence.
The answer runs to four paragraphs. It is a statement of position, not an announcement. It sets out what the Advertising Standards Authority of Singapore does under the Singapore Code of Advertising Practice, what the Codes of Practice issued under the Online Criminal Harms Act on 17 August 2026 require, which harms the Online Safety Commission can act on now, and that inauthentic material abuse is not among them yet.
It does not answer limbs (a) and (b) with a yes or a no. We state that as a fact about the document rather than reading an intention into the omission: the answer does not say that consent attestations will be required, and it does not say that persistent disclosure or expedited removal will be mandated.
The line worth quoting, and what it means for a victim today
The ministry said: "Inauthentic material abuse is one of the further categories of online harm under OSRAA that will be progressively implemented, as the OSC builds its capabilities in a sustainable and scalable way. We will provide more details on the commencement of this online harm for relief under OSRAA in due course."
Progressively implemented, in due course, and no date. For an adviser, that is the whole of the operative content. A person whose likeness or voice has been synthesised without consent cannot today bring that to the Online Safety Commission as inauthentic material abuse, because the category has not commenced.
The reason this is worth publishing is that the Act exists, the Commission exists, and the harm is named in the statute. The natural assumption is that the route is open. The ministry has said on the record that it is not.
The five harms that are live, and why AI-generated material can still fall within them
Relief under OSRAA, the ministry said, "currently covers the most prevalent and severe harms, namely intimate image abuse, image-based child abuse, doxxing, online harassment, and online stalking", adding that the Commission "can act on online material generated using AI and that meets the thresholds for these five harms".
That sentence does more work than the headline. None of the five is defined by how the material was produced. A synthetic intimate image may fall within intimate image abuse; a fabricated clip used to harass a person may be actionable as online harassment. The generation method is not the gate, the harm category is.
So the gap is specific, not general. What is missing is the category aimed at inauthenticity as such, which is the one that would reach an advertisement using a person's synthesised likeness to sell something without abusing, harassing or exposing them.
What covers an unauthorised likeness in an advertisement right now
On advertising standards, the ministry said ASAS applies the same approach regardless of whether AI is used, that advertisers are responsible for content being legal, decent, honest and truthful, that they "should disclose AI use where necessary to avoid misleading consumers", and that they "should not portray a person or their property without consent and ensure all endorsements or testimonials are genuine". ASAS takes a complaints-based approach and can get advertisers to amend or withdraw non-compliant advertisements.
Note the shape of that. It is a code applied after a complaint, with amendment or withdrawal as the outcome. The ministry describes no pre-publication consent attestation for AI-generated likeness, and we do not supply one.
On platforms, the ministry said the Government issued new and enhanced Codes of Practice under the Online Criminal Harms Act on 17 August 2026 to strengthen safeguards against online advertisements that impersonate individuals, and that designated online services "are required to verify advertisers' identities and prevent and promptly remove the publication of suspected scam advertisements". Those Codes predate our sweep window and we did not open them, so we report that requirement as the ministry describes it. On its own description, the duty is framed around suspected scam advertisements rather than around unauthorised likeness generally.
What we did not verify
What we opened: the MDDI newsroom page carrying the answer, retrieved and read in full, including the question as put, the four paragraphs of the answer, and the 10 September 2026 dateline and the Parliament Sitting header. Every quotation in this piece is from that page.
What we did not open: the Online Safety (Relief and Accountability) Act itself and its commencement notifications, the Online Criminal Harms Act Codes of Practice of 17 August 2026, the Singapore Code of Advertising Practice, and the Hansard record of the sitting. We therefore do not state the statutory definition of inauthentic material abuse, we do not state which provisions of OSRAA are or are not in operation beyond what the ministry said, and we describe the Codes of Practice only as the ministry described them.
What we refuse to claim: we do not say Singapore has no protection against AI-generated likeness abuse, because the ministry identified five live harms the Commission can act on where AI-generated material meets their thresholds. We do not say a commencement date exists or is imminent, because the answer gives none and says details will come in due course. We do not say platforms will be required to obtain consent attestations or to apply persistent disclosure and expedited removal, because the answer does not address those two limbs of the question either way. We do not treat this answer as a rule, guidance or instrument, because it is none of those.
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.
If you advise on synthetic likeness in Singapore, stop routing the advice through inauthentic material abuse. The ministry has said it is not commenced and has given no date. The practical route today is to ask whether the material meets the threshold of one of the five live harms, because those turn on what is done to the person rather than on how the material was made, and to treat the advertising side as a complaints-based code with amendment or withdrawal as the remedy.
Source File
Open the MDDI newsroom page and confirm three things: the three limbs of Question No. 29 as put, the sentence listing the five harms the Online Safety Commission currently covers and the statement that it can act on AI-generated material meeting their thresholds, and the final paragraph on inauthentic material abuse being progressively implemented with details in due course.
Inauthentic material abuse is one of the further categories of online harm under OSRAA that will be progressively implemented, as the OSC builds its capabilities in a sustainable and scalable way. We will provide more details on the commencement of this online harm for relief under OSRAA in due course. ยท MDDI, written answer to Parliamentary Question No. 29, final paragraph, 10 September 2026
FAQ
Is this a new rule?
No. It is a ministerial written answer to a parliamentary question. It binds nobody, creates no obligation and changes no law. Its value is as the government's own statement of where the commencement position stands.
Can the Online Safety Commission act on a deepfake today?
It depends on the harm rather than on the technology. The ministry said the Commission can act on AI-generated material that meets the thresholds for intimate image abuse, image-based child abuse, doxxing, online harassment or online stalking. The separate category of inauthentic material abuse has not commenced.
When will inauthentic material abuse commence?
The answer gives no date. It says the category will be progressively implemented as the Commission builds its capabilities, and that more details on commencement will be provided in due course.
Does an advertiser have to get consent before using an AI-generated likeness?
The answer does not impose a pre-publication attestation. It says advertisers should not portray a person or their property without consent under the Singapore Code of Advertising Practice, which ASAS applies on a complaints basis and can enforce by having an advertisement amended or withdrawn.
Related briefings
Sponsored Training
Practical AI training for regulated professionals, built around verification, documentation and a defensible process. See the courses.