Proactive Evidentiary Communication in Singapore’s Information Environment

Complete paper

The complete supplied manuscript follows, including its case material, tables, method and references. The original formatted DOCX is available above.

Abstract

Governments sometimes need to communicate before a false or manipulative account becomes entrenched. Yet speed, repetition and literal truthfulness do not constitute an adequate doctrine. Many public controversies are not contests between truth and falsehood. They combine verifiable facts, selective framing, uncertain forecasts, distributional consequences and legitimate differences over values. Treating all of these as “disinformation” risks analytical error, political overreach and institutional loss of trust.

This paper proposes proactive evidentiary communication: timely, lawful, attributable and evidence-led communication that gives the public the material facts, context, uncertainty, trade-offs and reasoning needed to make an informed judgment. It uses two analytical axes. The first classifies the content as demonstrably false, materially incomplete, empirically uncertain or a legitimate policy disagreement. The second assesses how the content is distributed: organically, deceptively coordinated or linked to foreign interference. A third operational consideration—how much of the evidence can be made publicly inspectable—determines how confidently the state should speak.

The paper is organised around cases. Singapore examples include false claims during the COVID-19 XBB wave, the TraceTogether data-use controversy, National Service remuneration, foreign-linked social-media accounts and inauthentic “local” news websites, and election safeguards against realistic deepfakes. Overseas cases from France, Ukraine and the European Union are retained only where they illuminate a Singapore decision problem. Each case identifies what happened, how it should be classified, what the response did, what remains unproven, and the practical lesson for Singapore.

The recommended model is a proactive public-reason capability, not an information-offence organisation. Its purpose is to improve the public’s decision environment: establish what is known, identify what remains uncertain, expose material omissions, explain who benefits and who bears the cost, and distinguish legitimate disagreement from covert manipulation.

Keywords: strategic communication; disinformation; misinformation; public reasoning; government communication; Singapore; correction; foreign interference; trust; policy trade-offs

Executive Summary

Singapore operates in an information environment that combines high connectivity, relatively strong trust in established news brands, widespread use of social and messaging platforms, four official languages, close exposure to regional narratives, and a dense legal framework for online falsehoods, criminal harms, foreign interference and election integrity. These features provide strong response capacity. They also magnify the cost of a wrong classification or overconfident official claim.

The central recommendation is to adopt proactive evidentiary communication:

Timely, lawful and attributable communication that provides the material facts, context, uncertainty, trade-offs and reasoning required for the intended audience to make an informed judgment.

The doctrine begins with three questions.

1. What kind of content problem is this?

Class A — Demonstrably false fact: correct with accessible evidence.

Class B — Accurate but materially incomplete: supply the missing context without accusing the speaker of lying.

Class C — Uncertain empirical judgment: disclose assumptions, ranges, confidence and what would change the assessment.

Class D — Legitimate policy disagreement: state the competing objective fairly and explain the chosen balance of benefits, burdens and risks.

2. How is it being distributed?

Content accuracy and distribution behaviour are separate. A true fact may be amplified through covert foreign coordination; a false claim may spread organically through family chats. Assess separately whether the activity is organic, coordinated, deceptive or foreign-linked.

3. How much of the evidence is publicly visible?

Evidence may be fully inspectable, partly releasable or protected for operational reasons. The less evidence the public can inspect, the narrower and more qualified the public claim should be. Source protection cannot justify presenting an unverifiable conclusion as certainty.

The case evidence supports a selective rather than permanently offensive posture:

For Singapore, five adaptations are essential:

Design for closed messaging, not only public platforms. Corrections must be portable: a short message, a shareable visual and an inspectable evidence page.

Use language and messenger segmentation. Translation is not enough; examples, tone and trusted intermediaries should reflect the affected audience.

Exploit whole-of-government speed without suppressing dissent. Independent challenge should occur before high-impact publication or attribution.

Distinguish public communication from statutory enforcement. The same facts may support communication, platform action or legal directions, but these are different decisions with different thresholds.

Measure understanding and trust, not merely reach. A viral rebuttal that leaves citizens confused about the trade-off—or distrustful of the institution—is not a success.

The objective is not for the Government’s version to be heard first. It is for citizens to end up with a more accurate and complete account of what is known, what is uncertain, who gains, who pays, what alternatives exist and why the decision was made.

1. Why a Singapore-Specific Doctrine Is Necessary

International frameworks are useful, but they cannot simply be imported. Singapore’s information environment changes the operating assumptions behind many overseas cases.

1.1 High connectivity and fragmented attention coexist

The Reuters Institute’s 2026 Singapore survey reports 94% internet penetration and overall trust in news at 46%, above its stated global average. Established brands such as CNA and The Straits Times continue to score highly, while audiences increasingly encounter news through digital platforms, social media and AI-mediated interfaces.[1] These findings should be read as survey indicators rather than a complete map of public opinion, but they capture an important duality: Singapore retains strong institutional channels while attention is increasingly fragmented.

This means the Government can often place an authoritative account into the public domain quickly. It cannot assume that the account will reach the people who saw the original claim, appear in the same format or be trusted by every segment.

1.2 Closed messaging is a distinct operational problem

During COVID-19, the Gov.sg WhatsApp channel grew from about 7,000 to more than 900,000 subscribers in ten weeks and distributed updates in the four official languages.[2] This demonstrated the value of direct, portable and multilingual communication.

The same environment allows questionable material to circulate through trusted family, workplace, religious and community networks. Small Singapore studies have found that respondents perceived false information as especially prevalent in instant-messaging settings and that a fact-checking chatbot could help users assess forwarded material, although the studies’ sample sizes and experimental designs do not justify population-wide claims.[3][4] The practical implication is sound even if the magnitude remains uncertain: a correction strategy designed only for Facebook, news conferences or a government webpage will miss part of the information flow.

1.3 Singapore’s audiences overlap

A message intended for a domestic audience can be read in Malaysia, Indonesia, China, India, the wider region and diaspora networks within minutes. English, Mandarin, Malay and Tamil versions may carry different cultural associations. Foreign-policy narratives can be reframed through ethnic, religious or civilisational identities. MHA has stated that a 2016–2017 hostile information campaign used Mandarin content and chat applications to target Singaporeans during a period of bilateral tension.[5]

The implication is not that every cross-border narrative is foreign interference. It is that communicators should assess domestic meaning, regional interpretation and diplomatic consequence together.

1.4 Whole-of-government coordination is both an advantage and a risk

Singapore can assemble evidence, policy expertise, legal advice and distribution channels quickly. The COVID-19 response showed the value of Gov.sg channels, Factually, agency websites, chat platforms and media briefings working together.[2]

The same concentration can propagate a mistaken assumption across agencies. A common clearance chain may create consistency without genuine challenge. The doctrine must therefore preserve speed while introducing a deliberate point at which someone can ask: Are we classifying this correctly? What would a sceptical but reasonable citizen think is missing?

1.5 Legal authority and communication credibility interact

Singapore has distinct statutory frameworks for false statements of fact, hostile foreign information campaigns, online criminal harms and realistic manipulated election advertising.[6][7][8][9] These frameworks are not interchangeable. A public correction, a platform direction, an account restriction and a criminal investigation serve different purposes and operate at different thresholds.

Even where a legal test is satisfied, the communication question remains: has the public been shown enough to understand the decision, its limits and the route for correction or challenge? State communications will be interpreted in the context of the state’s wider information-governance record. Legal power can authorise intervention; it cannot by itself generate trust.

2. The Doctrine: Three Axes Before Action

2.1 Axis One — Classify the content

Class A — Demonstrably false factual claim.
The proposition can be tested against reliable evidence: a fabricated government notice, invented quotation, false claim that a law has taken effect, or fake casualty report. The response is concise correction, proportionate to reach and harm, with evidence the audience can inspect.

Class B — Accurate but materially incomplete claim.
The stated fact may be correct, but missing information would substantially alter a reasonable person’s interpretation. Examples include announcing a policy benefit without its eligibility limits, citing gross cost without offsets, or presenting a lawful power without disclosing a material exception to an earlier assurance. The response is contextualisation and acknowledgement, not accusation.

Class C — Uncertain empirical judgment.
The dispute turns on a forecast, model, causal inference or evolving evidence: disease severity, deterrence effects, fiscal projections or behavioural response. The response should show assumptions, ranges, confidence and the conditions under which the assessment would change.

Class D — Legitimate policy disagreement.
The parties may agree on the main facts but give different weight to fairness, security, freedom, cost, solidarity or individual responsibility. The response is public argument: state the competing objective fairly, explain the Government’s choice and disclose the price of that choice.

2.2 Axis Two — Assess distribution and provenance

Ask separately:

Is the spread organic, coordinated or artificially amplified?

Are accounts authentic?

Is authorship attributable?

Are local identities, media brands or community voices being imitated?

Is there evidence of foreign direction, funding or infrastructure?

A campaign may contain true, false and opinion content at the same time. Foreign origin does not make every sentence false; factual accuracy does not make covert interference acceptable.

2.3 Axis Three — Grade evidence visibility

Level 1 — Publicly inspectable evidence.
Records, data, images, legislation or source documents can be released. The institution may speak with specificity, subject to ordinary uncertainty.

Level 2 — Partly releasable evidence.
Some indicators, methodology or corroborating material can be disclosed, but sources or capabilities require protection. The claim should identify what is evidence, what is inference and what remains withheld.

Level 3 — Protected evidence.
The core basis cannot be exposed without serious operational cost. The public claim should be narrower, confidence should be stated, and independent or statutory review should carry more weight. The institution should not ask the public to treat an uninspectable conclusion as self-proving.

This third axis addresses a recurring problem in national-security communication: what should the state say when the evidence is real but the public cannot see it? The answer is not silence in every case. It is bounded specificity, explicit confidence, disclosure of safe indicators, and a credible process of independent challenge.

3. What the Research Supports—and Does Not

3.1 Corrections generally reduce false belief

Meta-analytic and cross-national research indicates that factual corrections usually improve accuracy, although they may not erase all continuing influence from the original claim.[10][11] Strong fears that correction routinely entrenches the falsehood—the familiar “backfire effect”—are not supported as a general rule.[12][13]

The implication is straightforward: fear of backfire should not prevent correction of a significant Class A claim. It does not follow that every false statement deserves an official response. Response decisions still require assessment of reach, harm, amplification risk and messenger credibility.

3.2 Pre-bunking has bounded value

Technique-based inoculation can help people recognise manipulation devices such as false dichotomies, scapegoating and emotionally manipulative framing.[14] More recent longitudinal and feed-based studies show that effects decay, require reinforcement and may weaken sharply when interventions compete with ordinary content in realistic feeds.[15][16]

Pre-bunking is therefore useful for predictable techniques and recurring scams. It should not be presented as a population-level shield or as evidence that authorities can “immunise” citizens against manipulation.

3.3 Transparency is not a magic trust intervention

Evidence-based explanation and disclosure can support trust, especially when they help people understand decisions and performance.[17] However, transparency effects are conditional. Simply releasing more material does not automatically increase confidence, and disclosure can expose poor performance or create new questions.[18][19]

The relevant standard is not maximum disclosure. It is decision-useful transparency: provide what people need to understand the action, reduce material uncertainty, identify limits and hold the institution accountable.

3.4 Uncertainty should be communicated, not hidden

Uncertainty information can improve the public’s ability to judge evidence when it is expressed in usable terms.[20] The practical task is to avoid two failures: overstating confidence in order to sound decisive, or presenting so many caveats that the audience cannot act.

A useful form is:

What we know. What we do not yet know. What we are doing now. What would change our assessment. When we will update.

3.5 Government campaigns are weakly evaluated

Many studies measure recognition, perceived accuracy or stated sharing intention. Far fewer establish durable behavioural change, social cohesion or trust. Government case studies often report speed, output and reach rather than counterfactual impact.

This paper therefore distinguishes three levels of claim:

Observed: what the institution did and what immediately happened.

Plausible contribution: a reasonable inference supported by timing and mechanism.

Demonstrated effect: an outcome supported by credible evaluation or counterfactual evidence.

Most strategic-communication cases reach the first or second level, not the third.

4. Singapore Case Studies

Case 1 — XBB Claims During the 2022 COVID-19 Wave

What happened

In October 2022, claims circulated that Singapore was experiencing a severe and deadly wave caused by the XBB subvariant, including assertions about large increases in severe illness and deaths. MOH stated that these claims were false, reported that local data showed a lower hospitalisation risk for XBB cases than for BA.5 cases at that point, and issued POFMA correction directions to the originating publication and several republishers.[21]

Classification

Claims of a large, concealed increase in deaths or severe disease: Class A, if directly contradicted by official health records.

Assessment of XBB’s future severity or vaccine impact: Class C, because evidence was evolving.

Spread through private chats and reposting: a trajectory and channel problem, not proof of coordination.

What the response did well

The clarification identified specific false propositions and supplied contemporaneous local data. It acted through the same platforms on which the material was circulating and preserved access to the original content while appending a correction, consistent with the basic POFMA correction model.[6]

What remains unproven

Publicly available material does not establish how many people changed their beliefs, whether the intervention reduced forwarding in closed chats, or whether the official response amplified the claim among previously unexposed audiences.

Singapore lesson

A health correction should not collapse a false claim and a genuinely uncertain scientific question into the same sentence. A stronger response architecture would contain:

The false claim: stated once and precisely.

The current evidence: local data and its date range.

The uncertainty: what scientists were still assessing.

The action: what the public should do now.

The update point: when new evidence would be published.

For closed messaging, the response also needs a forwardable card in each relevant language and a link to a stable evidence page. The objective is not only to rebut; it is to give a recipient something credible and socially easy to send back into the same chat.

Case 2 — TraceTogether and the Limits of Literal Accuracy

What happened

Early official communications emphasised that TraceTogether data was collected for contact tracing. A May 2020 factsheet stated that data transferred to MOH would be accessed by authorised personnel for contact-tracing purposes.[22] On 4 January 2021, the Government clarified in Parliament that the Police could obtain TraceTogether data under the Criminal Procedure Code for criminal investigations.[23] The minister responsible for the Smart Nation initiative said he had not considered the CPC when speaking earlier.[24] On 8 January, the Government expressly acknowledged its error in not stating that TraceTogether data was not exempt from the CPC and announced legislation restricting non-contact-tracing use to serious offences.[25]

Classification

This was not primarily external misinformation. It was an institutional Class B failure: earlier statements created a materially incomplete public impression about purpose limitation, even if the legal power existed throughout.

What the response did well

The Government acknowledged error, explained the legal position, narrowed future use through legislation and later reported that police had requested TraceTogether data only once, for a murder investigation in May 2020.[26]

What remains unproven

There is no complete public retrospective showing the controversy’s effect on adoption, willingness to participate in future public-health technologies or segment-level trust. In May 2026, MOH said detailed programme records had been deleted when the programme stood down, limiting retrospective cost-effectiveness evaluation.[27]

Singapore lesson

The case demonstrates why “every sentence was technically true” is an inadequate defence. Public understanding is shaped by the whole representation, including emphatic assurances, omissions and reasonable inferences.

For programmes dependent on public participation, communicators should publish a purpose-and-exceptions statement before launch:

primary purpose;

data collected and not collected;

who may access it;

legal exceptions;

retention and deletion;

oversight and audit;

what happens if the policy changes.

When an earlier assurance proves incomplete, the correction should not be buried as a technical clarification. It should state plainly: what we previously said, what was missing, why it matters, what has changed, and how future recurrence will be prevented.

TraceTogether is therefore a central case for this doctrine. Proactive evidentiary communication must apply to the Government’s own claims, not only to claims it wishes to rebut.

Case 3 — National Service Remuneration as Policy Disagreement

What happened

On 29 February 2024, Workers’ Party MP Gerald Giam proposed that NSF payments be treated more like remuneration, including reference to the Local Qualifying Salary and CPF contributions. Senior Minister of State Heng Chee How replied that National Service is a duty rather than employment and that the allowance supports basic personal upkeep, while other needs are provided.[28][29]

MINDEF announced on 3 March 2025 that monthly NSF allowances would increase by S75 across ranks from 1 July 2025.[30] On 27 February 2026, Workers’ Party MP Kenneth Tiong renewed the proposal for second-year NSFs with reference to the S$1,800 Local Qualifying Salary. Senior Minister of State Zaqy Mohamad distinguished the LQS as an employment-related administrative tool from NS as civic duty, while acknowledging opportunity cost and explaining the existing support model.[31][32]

Classification

The debate combines Class C and Class D, with Class B risks on both sides.

How should society value sacrifice and opportunity cost? D

Should payment reflect upkeep, contribution, forgone income or need? D

Would salary treatment create CPF, legal or institutional consequences? C/D

What is the full recurrent cost and eligible cohort? C

What alternative benefits or public spending would be forgone? C/D

Would market-style remuneration strengthen fairness or alter the meaning of universal service? C/D

What the responses did well

Both sides articulated a legitimate organising principle: opportunity cost and fair recognition on one side; civic duty and non-employment status on the other. MINDEF also made actual allowance changes, showing that “duty” does not imply frozen support.

What remains incomplete

The public record does not place all major options into one comparable frame. A citizen cannot easily see the cost, distributional effect, CPF implications, targeted-support alternative and opportunity cost of each option in one place.

Singapore lesson

This is not a case for “countering disinformation”. It is a case for trade-off communication.

A complete public explanation should set out:

The Government need not accept the opposing proposal. It should engage the strongest version of it and explain why its preferred balance is better. The correct communications objective is trade-off comprehension, not narrative victory.

Case 4 — Foreign-Linked Accounts and Inauthentic “Singapore” Websites

What happened

In July 2024, MHA directed five social-media platforms to restrict 95 accounts linked to a foreign network. Ninety-two accounts had published more than 120 coordinated posts about Singapore’s leadership transition, alleging foreign control over the selection of Singapore’s fourth-generation leader. MHA later said the local reach was low.[33][34]

In October 2024, Singapore blocked ten inauthentic websites that used Singapore-associated names and visual features, carried local content—some AI-generated—and were assessed as potential platforms for hostile information campaigns.[35] In April 2026, six further sites were blocked. Four had carried GE2025-related material during the campaign period after being largely inactive before the Writ; several reproduced material from Singapore and foreign news outlets while presenting it as their own.[36]

Classification

This case sits primarily on the distribution and provenance axis:

Some claims may be Class A, B or D in substance.

Coordinated accounts, deceptive local branding and foreign-linked infrastructure are separate indicators of manipulation.

Current reach, latent capability and actual harm are distinct propositions.

What the response did well

The public statements described observable techniques: coordinated posting, dormant or suddenly active infrastructure, local-brand mimicry, content laundering and links to wider networks identified by external threat researchers. The Government also distinguished low current reach from potential future use.

What remains difficult

The public cannot independently inspect all evidence of foreign operation, network linkage or intent. In September 2025, MHA stated that it would not disclose detailed statistics or operational information about foreign online influence operations.[37] This creates an unavoidable evidence-asymmetry problem.

Singapore lesson

The state should use a graduated public-evidence ladder:

Observable facts: domains, posting dates, shared content, account behaviour.

Analytical indicators: coordination patterns, infrastructure linkage, mimicry techniques.

Attribution assessment: foreign linkage with a stated confidence level.

Intent assessment: potential or actual hostile information campaign.

Protected basis: what cannot be released and why.

The public statement should avoid sliding from one rung to the next. “These sites are inauthentic and linked to a known foreign network” is different from “a named state directed an active campaign that changed Singaporean opinion”.

Where current reach is low and action is anticipatory, the Government should say so. Pre-emption may be justified, but its success claim should be correspondingly modest: a potential vector was disrupted, not public opinion was protected unless outcome evidence exists.

The case also suggests a practical resilience measure: maintain a publicly searchable registry of known inauthentic local-looking domains, screenshots of their mimicry techniques, and guidance for newsrooms, community leaders and citizens. This teaches the public what to look for without requiring disclosure of sensitive sources.

Case 5 — Election Safeguards Against Realistic Deepfakes

What happened

Singapore amended its election laws in 2024 to prohibit, during the election period, digitally generated or manipulated online election advertising that realistically depicts a candidate saying or doing something the candidate did not say or do. The regime applies to sharing and reposting, allows affected candidates to request review by the Returning Officer, and penalises false or misleading declarations by candidates.[8][38] The safeguards operated for the 2025 General Election.[39]

Classification

This is not a response to one proven incident. It is anticipatory risk governance for a narrow Class A category: realistic false representation of a candidate’s speech or conduct.

What the design did well

The prohibition is limited by time, subject, realism and type of content. It does not purport to determine whether a candidate’s opinion, promise or interpretation is “misleading”. It also places responsibility on the depicted candidate to make a formal declaration.

What remains uncertain

There is limited public evidence about deterrent effect, false positives, speed of detection, vulnerable-group comprehension or how the regime would perform during a high-volume coordinated attack. MDDI has said deepfake detection tools evolve and does not publish their accuracy levels.[40]

Singapore lesson

The design illustrates a broader principle: narrow, objectively testable interventions are more defensible than broad authority over political “misinformation”.

Operationally, however, law is only one layer. A deepfake response plan should include:

authenticated original video or audio;

a rapid public denial by the depicted person;

forensic assessment expressed with confidence, not technological mystique;

platform escalation;

a media briefing package;

vernacular shareables;

an archive preserving the fake for evidence while limiting further amplification; and

a post-event review of whether the denial reached exposed audiences.

The response must also distinguish malicious realistic manipulation from obvious satire, parody, dubbing and ordinary editing. Overbroad enforcement would create the classification failure the doctrine is designed to prevent.

5. Overseas Cases—Used Through a Singapore Lens

Case 6 — Macron Leaks: Preparation Made Restraint Possible

What happened

Hacked material from Emmanuel Macron’s campaign was released shortly before the 2017 French presidential-election silence period. The campaign issued a brief factual statement; the media regulator reminded broadcasters and online outlets of legal responsibilities; and major French media exercised restraint. The material gained limited domestic traction.[41]

What remains unproven

No single intervention “defeated” the operation. Timing, media norms, election law, prior preparation, the quality of the material and attacker limitations all contributed.

Singapore adaptation

Singapore should not import France’s media and legal assumptions. It should import the operating principle: prepare early enough that leaders can choose a small response.

For a Singapore election or communal incident, preparation should include pre-cleared authenticity channels, platform contacts, multilingual holding lines, a candidate or agency verification protocol, and criteria for when silence protects the public better than rebuttal. Without preparation, organisations often over-respond because they have no safer option.

Case 7 — Ukraine Disclosures: Speaking When the Public Cannot See Everything

What happened

Before Russia’s full-scale invasion of Ukraine in February 2022, the United States and United Kingdom released intelligence about possible false-flag activity and Russian political plans at unusual speed. Subsequent events broadly validated the central warning that invasion was imminent.[42][43]

What the case supports

Selective disclosure can deny an adversary uncontested narrative space, establish a common warning picture and make later fabrications less credible.

What remains unproven

Claims that disclosure caused allied unity, disrupted specific Russian plans or deterred actions are plausible but not demonstrated counterfactuals. The invasion was not deterred.[44][45]

Singapore adaptation

For a small state, premature or overstated attribution can impose high diplomatic and credibility costs. Intelligence-based communication should therefore use a release test:

Is the anticipated harm serious enough?

Is the evidence strong and independently challenged?

What indicators can safely be shown?

Can the claim be narrowed to what is actually supported?

What confidence level should be stated?

What happens if new evidence contradicts the assessment?

Where evidence is protected, Singapore should prefer bounded warning over totalising accusation. It may be defensible to say that a cluster of accounts is coordinated and foreign-linked without claiming that a named government directed every message or achieved measurable influence.

Case 8 — EUvsDisinfo: One Misclassification Can Define the Programme

What happened

In 2018, EUvsDisinfo removed entries concerning three Dutch media articles after complaints. The outlets pursued legal action and the Dutch parliament passed a motion seeking the project’s abolition.[46][47]

Singapore adaptation

Singapore’s concentrated institutional environment makes the lesson especially important. A prominent error by one government-linked fact-checking or counter-interference function may affect perceptions of the wider system, including unrelated agencies and future corrections.

Publicly labelling a claim, outlet or actor should therefore require:

a clear statement of the exact proposition being assessed;

distinction between false fact, omission, opinion and poor journalism;

evidence available to the subject before publication where feasible;

review by someone outside the initiating chain;

a visible appeal and correction route; and

equal-prominence correction when the institution is wrong.

A correction log is necessary but insufficient. The stronger safeguard is preventing the wrong category from being applied in the first place.

6. A Singapore Operating Model

Step 1 — Write the claim in testable form

Do not respond to a mood, hashtag or broad “narrative”. Write the proposition that is said to be false, incomplete or harmful. Separate multiple propositions.

Step 2 — Classify on both axes

Record:

content class: A, B, C or D;

distribution: organic, coordinated, deceptive or foreign-linked;

evidence visibility: Level 1, 2 or 3.

A response should not proceed at high intensity until these judgments and their basis are recorded.

Step 3 — Assess harm, trajectory and audience

Consider:

current reach and acceleration;

closed-chat circulation;

audience vulnerability;

racial, religious, security, diplomatic or electoral sensitivity;

whether authoritative information already exists;

whether a response will become the main amplifier;

whether the matter is likely to self-correct; and

whether delay creates irreversible harm.

Step 4 — Define the objective

Choose one primary objective:

correct a false fact;

fill a material information gap;

reduce uncertainty;

explain a policy trade-off;

expose a manipulation method;

maintain public safety;

preserve a record for accountability; or

disrupt a covert campaign.

“Win the narrative” is not measurable and encourages unnecessary escalation.

Step 5 — Choose the least escalatory effective response

no public response;

monitor and preserve evidence;

quietly brief journalists, platforms or community intermediaries;

improve the source webpage, dataset or FAQ;

issue a shareable clarification through subject-matter experts;

agency or ministerial correction;

expose coordinated behaviour;

make bounded foreign attribution;

take platform or statutory action under the appropriate framework.

Legal action and public communication may occur together, but one should not be used as a substitute for reasoning through the other.

Step 6 — Build the response in three layers

Layer 1 — Portable: one sentence, one visual, one action.
Designed for WhatsApp, Telegram, SMS, social media and community forwarding.

Layer 2 — Explainable: a short account of what is true, false, missing or uncertain.
Designed for news coverage, community leaders and interested citizens.

Layer 3 — Inspectable: source documents, data, timeline, methodology, legal basis and correction history.
Designed for scrutiny by journalists, experts, Parliament, courts and future reviewers.

Step 7 — Apply the completeness test

Before publication, ask:

Is every factual assertion supportable?

Have we omitted a fact that would materially change interpretation?

Have we separated evidence from inference?

Have we stated important uncertainty?

Have we represented the strongest opposing argument fairly?

Have we explained the material cost of the preferred course?

Have we distinguished public interest from partisan advantage?

Would we consider this fair if issued by a political opponent?

Step 8 — Select the messenger and language

The most authoritative source is not always the most persuasive messenger. Depending on the issue, use:

responsible minister or agency;

clinician, engineer, economist or operational expert;

community, religious or professional leader;

affected citizen or service user;

mainstream newsroom or independent fact-checker; or

a combination with clear role boundaries.

Translation should preserve meaning, but localisation should also address audience concerns, examples and cultural cues. A four-language release is not automatically a four-audience strategy.

Step 9 — Pre-commit to correction and update

Every major intervention should identify:

approving officer;

source record;

review time;

next-update commitment;

correction channel;

equal-prominence rule; and

pause, withdrawal or escalation triggers.

7. Institutional Design and Guardrails

7.1 Build a public-reason function, not an information-offence unit

Its role should be to classify, assemble evidence, expose uncertainty, explain trade-offs, coordinate channels, test amplification risk and recommend proportionate action. It should not determine which domestic political position is “true”.

7.2 Separate analysis, advocacy and enforcement

Analysis: Is the claim false, incomplete, uncertain or coordinated?

Advocacy: Why is the Government’s policy choice preferable?

Enforcement: Is a statutory test met, and what direction or sanction is lawful?

The same institution may contribute to more than one function, but the public product should not blur them.

7.3 Give red teams real authority

For sensitive interventions, reviewers should be able to:

demand the underlying evidence;

propose alternative explanations;

challenge legal and causal assumptions;

test the strongest opposing account;

delay publication; and

record unresolved disagreement.

For Level 3 protected evidence or named-state attribution, at least one reviewer should sit outside the initiating chain.

7.4 Use graduated attribution

Do not collapse these into one claim:

content is false;

content uses a manipulation technique;

accounts are coordinated;

activity is linked with stated confidence to a foreign network;

activity is directed by a foreign actor;

activity caused measurable harm.

Each requires a higher threshold.

7.5 Exercise election and domestic-contestation restraint

During elections and major domestic political disputes, official communication should focus especially tightly on verifiable administrative facts, public safety, voting integrity and continuity of essential services. Policy advocacy may be legitimate, but it should not be presented as neutral fact-checking.

7.6 Correct official errors visibly

A significant error should be corrected with prominence reasonably comparable to the original communication. The correction should preserve the record and state what changed and why. Silent webpage edits are insufficient where the original claim was widely distributed.

7.7 Create a case library

Singapore should maintain a practitioner case library containing:

original claim and channel;

classification on all three axes;

evidence available at the time;

chosen response and alternatives rejected;

outcome data;

what was later found to be wrong or incomplete;

trust and amplification effects; and

lessons for future playbooks.

The library should include official successes, non-intervention decisions and institutional mistakes. A system that studies only its “wins” will overestimate intervention value.

8. Measurement

A credible evaluation framework separates five outcomes.

8.1 Understanding

Can people state the core verified facts and distinguish them from what remains unknown?

8.2 Belief

Has acceptance of the specific false proposition changed among people exposed to it?

8.3 Trade-off comprehension

Can people identify both the main benefit and the main cost of the Government’s preferred policy and its principal alternative?

8.4 Behaviour

Has harmful forwarding, non-compliance, scam payment, hostility or other consequential behaviour changed?

8.5 Trust

Has confidence in the communicating institution strengthened or weakened, especially among sceptical and directly affected groups?

Reach, impressions, media pickup and response time are operating indicators. They are not outcomes.

Singapore-specific measurement requirements

measure by age, language and affected community, not only national average;

sample closed-messaging exposure through voluntary tiplines, panels or research partnerships rather than assuming public-platform data represents the whole environment;

measure whether portable corrections were forwarded into the same networks as the original claim;

track whether official intervention became the main source of awareness of a fringe claim;

repeat measures after meaningful intervals because correction and pre-bunking effects decay;

assess whether the audience can distinguish fact-checking from policy advocacy; and

publish post-event reviews for major interventions where security and privacy permit.

Conclusion

Proactive communication has a defensible role in countering falsehood and manipulation. But the doctrine must begin by recognising that many difficult public disputes are not contests between truth and lies.

When a claim is demonstrably false, government should correct it promptly, proportionately and with inspectable evidence. When a claim is accurate but incomplete, government should provide the missing context and apply the same completeness standard to its own communications. When the evidence is uncertain, government should disclose assumptions, ranges, confidence and what would change the assessment. When the disagreement concerns values and priorities, government should acknowledge the legitimate objective advanced by the other side, explain the benefits and burdens of each option, and state why it chose one balance over another.

Singapore’s advantages—coordination, capability, trusted channels and legal tools—make early action possible. They do not make early action automatically wise. In a small, multilingual and highly connected society, a major classification or attribution error can cross institutions and communities rapidly. The answer is not passivity. It is stronger diagnosis, material completeness, independent challenge, channel-aware delivery and modest claims about effect.

The standard is not whether the Government spoke first or dominated attention. It is whether citizens were left with a more accurate and complete understanding of:

what is known;

what remains uncertain;

what evidence they can inspect;

who benefits;

who bears the cost;

what alternatives exist;

whether distribution was covert or manipulated; and

why the decision was made.

A state protects credibility not merely by saying true things. It protects credibility by applying the same evidentiary discipline to itself, by resisting the temptation to classify disagreement as deception, and by trusting the public with the full policy choice.

Appendix A — Case Assessment Card

1. Incident
What exactly was published, by whom, where and when?

2. Testable propositions
Break the content into separate factual, inferential and value claims.

3. Classification
A, B, C or D; organic, coordinated or foreign-linked; evidence visibility Level 1, 2 or 3.

4. Harm and trajectory
Current reach, acceleration, affected audience, offline risk and amplification risk.

5. Response objective
Correction, context, uncertainty reduction, policy explanation, campaign disruption or public safety.

6. Options considered
Silence, indirect engagement, evidence publication, expert response, ministerial response, attribution or legal action.

7. Response package
Portable, explainable and inspectable layers; language and messenger plan.

8. What success means
Understanding, belief, trade-off comprehension, behaviour and trust.

9. Correction and update
Next update, correction route, withdrawal threshold and accountable officer.

10. Post-event finding
What worked, what remains unproven and what should change in the playbook?

Appendix B — One-Page Decision Card

Classify

What exact proposition are we addressing?

Is it false, incomplete, uncertain or a value disagreement?

Is spread organic, coordinated or foreign-linked?

How much evidence can the public inspect?

Decide

What harm will occur if we do nothing?

Will our intervention amplify the claim?

What is the least escalatory effective response?

Are we communicating, advocating or enforcing—and have we kept these roles clear?

Draft

What is known?

What is unknown?

What action is being taken?

What material fact or cost might we be omitting?

Would this be fair if issued by a political opponent?

Deliver

Portable message and visual.

Short explanation.

Inspectable evidence page.

Appropriate language and messenger.

Credible next-update commitment.

Measure

Understanding.

Belief.

Trade-off comprehension.

Behaviour.

Trust.

Stop or correct

New evidence changes the classification.

The response is the main amplifier.

A factual error is found.

Protected evidence no longer supports the confidence claimed.

Trust falls materially in an affected group.

Method and Limitations

This paper is a practitioner-oriented synthesis of peer-reviewed research, official records, legislation, government statements and selected case analyses. Primary sources were preferred for event chronology and legal provisions. Official accounts establish what authorities said and did; they do not independently prove effectiveness or intent. Secondary and academic sources were used to provide comparative context and critique.

The case method improves practical relevance but carries selection risk. Cases with accessible records and visible controversy are easier to study than quiet non-interventions or failures that remain classified. Causal conclusions are therefore stated conservatively. “What happened” is distinguished from “what plausibly contributed” and “what was demonstrated”.

The Singapore audience data cited from the Reuters Institute is survey-based. The local instant-messaging studies are useful but do not constitute representative national prevalence estimates. Legal references are descriptive and do not constitute legal advice. Sources should be rechecked immediately before external publication.

References

Singapore information environment and legal framework

[1] Reuters Institute for the Study of Journalism. “Singapore.” Digital News Report 2026, 16 June 2026. https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/singapore

[2] Ministry of Communications and Information. “Gov.sg launches new channels to keep the public informed about COVID-19.” 2 April 2020. https://www.mddi.gov.sg/newsroom/gov-sg-launches-new-channels-to-keep-the-public-informed-about-covid-19/

[3] Lim, G., and Perrault, S. T. “Local Perceptions and Practices of News Sharing and Fake News.” 2020. https://arxiv.org/abs/2010.07607

[4] Lim, G., and Perrault, S. T. “Fact Checking Chatbot: A Misinformation Intervention for Instant Messaging Apps and an Analysis of Trust in the Fact Checkers.” 2024. https://arxiv.org/abs/2403.12913

[5] Ministry of Home Affairs. “Singapore and Foreign Interference.” Updated 30 January 2026. https://www.mha.gov.sg/what-we-do/foreign-interference/singapore/

[6] POFMA Office. “Home” and “Regulations.” Updated July 2026. https://www.pofmaoffice.gov.sg/

[7] Ministry of Home Affairs. “Introduction to Foreign Interference (Countermeasures) Act.” Updated 30 January 2026. https://www.mha.gov.sg/what-we-do/foreign-interference/fica/

[8] Ministry of Digital Development and Information. “New Legal Measures to Uphold Integrity of Online Advertising During Elections.” 9 September 2024. https://www.mddi.gov.sg/newsroom/new-legal-measures-uphold-integrity-online-advertising-elections/

[9] Ministry of Home Affairs. “Introduction of the Online Criminal Harms Bill.” 8 May 2023. https://www.mha.gov.sg/media-room/newsroom/introduction-of-the-online-criminal-harms-bill/

Behavioural and public-communication research

[10] Chan, M. S., Jones, C. R., Hall Jamieson, K., and Albarracín, D. “Debunking: A Meta-analysis of the Psychological Efficacy of Messages Countering Misinformation.” Psychological Science 28(11), 2017. https://doi.org/10.1177/0956797617714579

[11] Porter, E., and Wood, T. J. “The Global Effectiveness of Fact-checking.” PNAS 118(37), 2021. https://doi.org/10.1073/pnas.2104235118

[12] Ecker, U. K. H., Lewandowsky, S., and Chadwick, M. “Can Corrections Spread Misinformation to New Audiences?” Cognitive Research: Principles and Implications 5, 41, 2020. https://doi.org/10.1186/s41235-020-00241-6

[13] Swire-Thompson, B., DeGutis, J., and Lazer, D. “Searching for the Backfire Effect.” Journal of Applied Research in Memory and Cognition 9(3), 2020. https://doi.org/10.1016/j.jarmac.2020.06.006

[14] Roozenbeek, J., van der Linden, S., Goldberg, B., Rathje, S., and Lewandowsky, S. “Psychological Inoculation Improves Resilience against Misinformation on Social Media.” Science Advances 8(34), 2022. https://doi.org/10.1126/sciadv.abo6254

[15] Maertens, R., et al. “Psychological Booster Shots Targeting Memory Increase Long-term Resistance against Misinformation.” Nature Communications 16, 2025. https://doi.org/10.1038/s41467-025-57205-x

[16] “Limited Effectiveness of Psychological Inoculation against Misinformation in a Social Media Feed.” PNAS Nexus, 2025. https://doi.org/10.1093/pnasnexus/pgaf172

[17] OECD. Survey on Drivers of Trust in Public Institutions — 2024 Results. 2024. https://www.oecd.org/en/publications/2024/07/oecd-survey-on-drivers-of-trust-in-public-institutions-2024-results_eeb36452.html

[18] Grimmelikhuijsen, S. G., Piotrowski, S. J., and Van Ryzin, G. G. “Latent Transparency and Trust in Government.” Government Information Quarterly 37(4), 2020. https://doi.org/10.1016/j.giq.2020.101497

[19] Ripamonti, J. P. “Does Being Informed about Government Transparency Boost Trust?” Government Information Quarterly 41(3), 2024. https://doi.org/10.1016/j.giq.2024.101960

[20] Cruz-Cano, R., and Allison, D. B. “Report Uncertainty Information to Improve Trust in Science.” Nature Human Behaviour 9, 2025. https://doi.org/10.1038/s41562-024-02084-3

Singapore cases

[21] Ministry of Health. “Issuance of Correction Direction under POFMA on Falsehoods Regarding XBB Wave in Singapore.” 14 October 2022. https://www.moh.gov.sg/newsroom/issuance-of-correction-direction-under-the-protection-from-online-falsehoods-and-manipulation-act-on-falsehoods-regarding-xbb-wave-in-singapore/

[22] Smart Nation and Digital Government Office. “Factsheet — Enhancements to the TraceTogether Application.” 1 May 2020. https://www.mddi.gov.sg/newsroom/factsheet-enhancements-to-the-tracetogether-application/

[23] Ministry of Home Affairs. “Oral Reply to Parliamentary Question on Whether TraceTogether Data Will Be Used for Criminal Investigations.” 4 January 2021. https://www.mha.gov.sg/media-room/newsroom/oral-reply-to-parliamentary-question-on-whether-tracetogether-data-will-be-used-for-criminal-investigations-by-mr-desmond-tan-minister-of-state-ministry-of-home-affairs-and-ministry-of-sustainability-and-the-environment/

[24] Ministry of Communications and Information. “Clarification on the Usage of TraceTogether Data.” 5 January 2021. https://www.mddi.gov.sg/newsroom/20210105/

[25] Ministry of Communications and Information. “Upcoming Legislative Provisions for Usage of Data from Digital Contact Tracing Solutions.” 8 January 2021. https://www.mddi.gov.sg/newsroom/upcoming-legislative-provisions-for-usage-of-data-from-digital-contact-tracing-solutions/

[26] Ministry of Home Affairs. “Written Reply to Parliamentary Question on Use of TraceTogether Data for Criminal Investigations.” 4 April 2022. https://www.mha.gov.sg/media-room/newsroom/written-reply-to-pq-on-use-of-tracetogether-data-for-criminal-investigations

[27] Ministry of Health. “Assessing TraceTogether’s Contribution to COVID-19 Contact Tracing, Total Programme Cost and Lessons for Future Digital Tracing.” 7 May 2026. https://www.moh.gov.sg/newsroom/assessing-tracetogether-s-contribution-to-covid-19-contact-tracing--total-programme-cost-and-lessons-for-future-digital-tracing-/

[28] Giam, G. “National Service Pay.” 29 February 2024. https://geraldgiam.sg/2024/02/ns-pay/

[29] Ministry of Defence. “Speech by Senior Minister of State for Defence, Mr Heng Chee How, at the Committee of Supply Debates 2024.” 29 February 2024. https://www.mindef.gov.sg/news-and-events/latest-releases/29feb24_speech/

[30] Ministry of Defence. “Fact Sheet: Updates to Initiatives for NSFs and NSmen.” 3 March 2025. https://www.mindef.gov.sg/news-and-events/latest-releases/03mar25_fs3/

[31] Workers’ Party. “Second-year NSF Allowance — S$1,800 Local Qualifying Salary.” 27 February 2026. https://www.wp.sg/parliament/speech-by-kenneth-tiong-on-mindef-second-year-nsf-allowance---1-800-local-qualifying-salary-lqs

[32] Ministry of Defence. “Speech by Senior Minister of State for Defence, Mr Zaqy Mohamad, at the Committee of Supply Debate 2026.” 27 February 2026. https://www.mindef.gov.sg/news-and-events/latest-releases/27feb26-speech1/

[33] Ministry of Home Affairs. “Issuance of Account Restriction Directions under FICA against 95 Social Media Accounts Linked to a Foreign Network.” 19 July 2024. https://www.mha.gov.sg/media-room/newsroom/issuance-of-account-restriction-directions-under-the-foreign-interference-countermeasures-act-2021-against-95-social-media-accounts-linked-to-a-foreign-network/

[34] Ministry of Home Affairs. “Oral Reply on Details Regarding Posts Made by the 95 Social Media Accounts.” 7 August 2024. https://www.mha.gov.sg/media-room/newsroom/oral-reply-to-parliamentary-question-on-details-regarding-posts-made-by-the-95-social-media-accounts-that-were-blocked-on-19-july-2024/

[35] Ministry of Home Affairs. “Ten Inauthentic Websites Blocked for Potential Threat of Being Used to Mount Hostile Information Campaigns against Singapore.” 22 October 2024. https://www.mha.gov.sg/media-room/newsroom/ten-inauthentic-websites-blocked-for-potential-threat-of-being-used-to-mount-hostile-information-campaigns-against-singapore/

[36] Ministry of Home Affairs. “Six Inauthentic Websites Blocked for Potential Threat of Being Used to Mount Hostile Information Campaigns against Singapore.” 23 April 2026. https://www.mha.gov.sg/media-room/newsroom/six-inauthentic-websites-blocked-for-potential-threat-of-being-used-to-mount-hostile-information-campaigns-against-singapore/

[37] Ministry of Home Affairs. “Number of Foreign Online Influence Operations Detected in the Past Three Years.” 23 September 2025. https://www.mha.gov.sg/media-room/newsroom/number-of-foreign-online-influence-operations-detected-in-the-past-three-years/

[38] Ministry of Digital Development and Information. “Opening Speech at the Second Reading of the ELIONA Bill.” 15 October 2024. https://www.mddi.gov.sg/newsroom/opening-speech-by-minister-josephine-teo-at-the-second-reading-of-the-eliona-bill/

[39] Elections Department Singapore. “Online Campaigning.” Updated 2025. https://www.eld.gov.sg/candidate_parliamentary_campaign_online.html

[40] Ministry of Digital Development and Information. “Response to Parliamentary Question on Accuracy of Deepfake Detection Technologies.” 7 August 2024. https://www.mddi.gov.sg/newsroom/mddi-s-response-to-pq-on-accuracy-of-deepfake-detection-technologies/

Overseas comparative cases

[41] Vilmer, J.-B. J. The ‘Macron Leaks’ Operation: A Post-Mortem. Atlantic Council, 2019. https://www.atlanticcouncil.org/in-depth-research-reports/report/the-macron-leaks-operation-a-post-mortem/

[42] UK Foreign, Commonwealth & Development Office. “Kremlin Plan to Install Pro-Russian Leadership in Ukraine Exposed.” 22 January 2022. https://www.gov.uk/government/news/kremlin-plan-to-install-pro-russian-leadership-in-ukraine-exposed

[43] Central Intelligence Agency. “DCIA Remarks at Truman Library Institute Award Dinner.” 29 April 2022. https://www.cia.gov/stories/story/truman-award-dcia-remarks/

[44] US Naval Institute. “The Promise and Danger of Declassifying Intelligence for Effect.” Proceedings, April 2023. https://www.usni.org/magazines/proceedings/2023/april/promise-and-danger-declassifying-intelligence-effect

[45] Centre for International Governance Innovation. “Deterrence, Disruption and Declassification: Intelligence in the Ukraine Conflict.” 2 May 2022. https://www.cigionline.org/articles/deterrence-disruption-and-declassification-intelligence-in-the-ukraine-conflict/

[46] EUvsDisinfo. “Removal of Three Cases Further to Complaints by Dutch Media.” 2018. https://euvsdisinfo.eu/removal-of-three-cases-further-to-complaints-by-dutch-media/

[47] Poynter. “Three Publications Are Suing the EU over Fake News Allegations.” 28 February 2018. https://www.poynter.org/fact-checking/2018/three-publications-are-suing-the-eu-over-fake-news-allegations/

All online sources were checked during the revision completed on 28 July 2026. Legal, policy and current-affairs sources should be rechecked immediately before external circulation.