On 4 August 2026, Singapore's Ministry of Home Affairs published Senior Minister K Shanmugam's parliamentary reply on enforcing the Online Criminal Harms Act (OCHA) against non-English overseas platforms. He said the Act applies to "all online platforms accessible to Singapore users, independent of language medium," and covers AI-generated content. Directions were issued to "a number of platforms" and, he said, "they have complied." The published reply does not name the platforms or set out the exact orders. It also does not say whether access to any service was blocked.
The strongest case for the government
The government's position deserves a fair hearing. Small, multi-ethnic city-states have a real interest in stopping content that incites racial hatred, and the harm can spread faster than a court can sit. The Home Affairs Ministry says it acted on 14 posts that the Singapore Police Force assessed as likely to breach Section 298A of the Penal Code, which covers promoting enmity between racial groups. The platforms complied, so the system worked without escalation. Publishing a list of the platforms or the orders could also hand a roadmap to the actors the state is trying to counter. And the Act was designed so that blocking is the last step, not the first.
That last point matters. According to Allen & Gledhill's summary of the Act as passed on 5 July 2023, Access Blocking, App Removal and Service Restriction Orders may be issued "in the event of non-compliance with a Direction, a Rectification Notice, an Implementation Directive, or another Order." The same summary says the Act provides appeal mechanisms against Directions, orders issued for non-compliance, and other measures. The Bill as introduced is the underlying text. I could not extract the Bill's section-level provisions, so I rely on the secondary summary for them.
Where the transparency gap bites
The case for secrecy is weaker than it looks, and the government's own conduct shows why. When it acted on the 14 posts in June, it publicly named YouTube, Facebook and X, the number of posts, the statutory basis and the issuing authority. The August reply on non-English platforms has none of that. The difference may be operationally justified, because smaller overseas services may be less willing to be named. But the public cannot tell from the record, and that is the problem.
Three specific gaps follow.
- No public record of recipients. A reader cannot tell whether the platforms are large or small, or whether they host political speech, diaspora communities or purely criminal content. Without that, "they complied" cannot be checked.
- No visibility of the escalation ladder. The reply says further action may include prosecution and "the issuance of orders to restrict Singaporean access to the non-compliant platform." A Direction is a request to disable specific content. A blocking order cuts off an entire service for every user in the country. The two carry very different speech costs, yet the public has no data on how often the second has been used.
- Appeals exist on paper but are hard to use blind. An appeal right is only meaningful if the affected party, or the users who lost access, can learn an order exists. The June notice did not describe a review route for affected content. A statutory appeal that nobody knows about protects few people.
Why proportionality needs data, not trust
Proportionality is a test of evidence: whether the measure was suitable, whether a narrower one would have done, and what it cost. Blocking orders are blunt because ISPs and app stores usually cannot remove one post. They can only remove a domain, an app or a service tier. Over-blocking is the standard failure mode of that tool. Measuring it requires publishing at least counts of orders, the category of offence, the type of order, and how long each lasted.
A transparency regime does not need to expose targets mid-operation. Many jurisdictions publish aggregate reports with a delay. A similar Singapore report could cover the number of Directions by offence category, the number of escalations to Access Blocking, App Removal or Service Restriction Orders, appeal outcomes, and whether orders were later lifted. Individual platform names could be withheld for a fixed period where there is a specific security reason, with the reason recorded and reviewed.
This matters beyond Singapore. Governments use language about "social cohesion" and "disinformation" to justify access restrictions. Civil society groups such as Access Now, which has run its #KeepItOn campaign against shutdowns for a decade, argue that blocking powers are least harmful where their use is documented and contestable. Singapore's reputation as a rules-based, investment-friendly hub is an asset. Regional platforms and cloud providers deciding where to locate staff and servers will look for predictability. Opaque powers are a cost on that predictability even if they are rarely used.
A practical standard
The proportionate answer is not to remove the powers. It is to attach reporting to them. Four steps would do it:
- A periodic public report on OCHA Directions and Orders by type and offence category.
- A published rule on when platform names will and will not be disclosed.
- Notice to the public, such as a landing page on a blocked service, whenever an Access Blocking Order takes effect.
- Independent review data on appeal outcomes.
None of these slow the removal of genuinely criminal content. They let Singaporeans, platforms and outside observers confirm that a power built for narrow criminal harms is staying narrow. Until then, the August reply tells us that something was done, to someone, and that it worked. That is not enough to judge whether it was proportionate.