Govt to continue studying if levers beyond Pofma are needed to tackle deepfakes, online fake news

As deepfakes, non-factual claims and other online harms become more prevalent in cyberspace, the Republic is growing its capabilities to tackle them.

This involves a continual study of whether new levers are needed, said spokespeople for the Ministry of Communications and Information and Ministry of Law on June 16.

In particular, artificial intelligence (AI) is being misused to create and spread misinformation and disinformation, to manipulate public opinion and influence election results, they noted.

Responding to questions from The Straits Times on the effectiveness of Singapore’s fake news law, they added that the Protection from Online Falsehoods and Manipulation Act (Pofma) can be used to address such issues “under some conditions”, but did not elaborate.

In an article on June 15, ST reported that five years after Pofma came into force, concerns are beginning to emerge over the Act’s effectiveness amid the rise of AI, with generative-AI tools becoming easily accessible to the public. These include whether the law will be able to work quickly or precisely enough to address instances of AI-generated misinformation and disinformation.

ST had earlier sent questions on the issue to the authorities for the June 15 report, but they did not respond in time.

In their reply on June 16, the ministries’ spokespeople said Pofma is not the only tool the Government relies on to tackle fake news. For example, the Centre for Advanced Technologies in Online Safety works with various partners to build tools to detect and address various forms of online harms.

There are also public education efforts focused on helping people critically evaluate information and recognise false or misleading content.

In addition, the media and third-party fact-checkers play a key role in curbing the spread of misinformation, the spokespeople added.

Asked how Pofma has helped regulate Singapore’s fake news landscape over the past five years, they said: “Pofma has helped at least to address misinformation that undermines public interest. It establishes a factual basis for honest discourse in our public square.”

They added that it is essential to have measures that support the infrastructure of fact and promote honest speech in public discourse.

Under Pofma, the primary response deployed against falsehoods deemed as being against the public interest is correction directions. These do not require the falsehoods to be taken down, they said. Instead, a link to the facts is to be put up alongside the falsehood. “A discerning reader will be able to decide for himself or herself after reading both the falsehoods and facts,” said the spokespeople.

“Pofma has played an effective role in the upkeep of trust in our public institutions. In the long run, it will support good-quality public discourse and maintain trust in our public institutions,” they added.

In its June 15 report, ST had cited several experts concerned over how Pofma may some day be abused by individuals holding political authority, and who called for independent fact-checking platforms.

A spokesman for the Workers’ Party, whose MPs had all voted against the Pofma Bill in Parliament in 2019, said in an e-mail response late on June 14 that the party maintains that the courts must be required to approve government Pofma directives on substantive grounds.

“Should the courts be the primary arbiter of falsehoods, this will help to establish an independent, unbiased and equitable process that will be perceived by many to be more reliable, transparent and fair,” he said.

In response to a query on how ministers consider applications to vary or cancel Pofma directions, the ministries’ spokespeople said the minister issuing the direction assesses each application on a case-by-case basis.

Factors to consider include whether the recipient had communicated the statement in Singapore, if the statement is a false statement of fact, whether it is technically possible for the recipient to comply with the direction, and whether there is public interest in maintaining the direction.

If the minister refuses the application in whole or in part, the recipient of the Pofma direction can appeal to the High Court.

“The courts have the final say over whether the Pofma direction was correctly issued,” said the spokespeople.

Since the law came into force in 2019, there have been 17 such applications submitted to the relevant ministers.

Of the 14 applications to cancel Pofma directions, only one was successful.

This was the correction direction issued to the Australia-based East Asia Forum, which had carried an article written by a National University of Singapore academic.

The correction direction to the site and a targeted correction direction to Meta Platforms to carry a correction notice on Facebook were later cancelled. This was after the article was removed from all relevant sites at the request of the author, who also issued a public apology on the matter.

Of the 13 unsuccessful applications, eight recipients appealed to the High Court.

Of these, six were dismissed fully by the courts and one was withdrawn by the appellant.

One case was allowed in part by the courts – the one involving the Singapore Democratic Party that concluded in 2021 – and a fresh correction direction was subsequently issued to correct the falsehood, the spokespeople said.

The remaining three applications of the 17 submitted since 2019 were for variations, which were all allowed as the recipients of the directions were unable to comply for technical reasons.

Five years of Pofma: How has the law been used to combat fake news?

When a Facebook page made claims in April 2020 that the Singapore Government had covered up Covid-19 case numbers to reduce public panic, the authorities invoked the fake news law on the same day to rebut the falsehood.

It was one of the swiftest uses of the Protection from Online Falsehoods and Manipulation Act (Pofma) in the five years since the law came into force on Oct 2, 2019.

The Straits Times looked at every Pofma directive issued by the authorities since that date to understand what content might trigger a Pofma order, and how the Government might react to fake news from various sources – issues that saw robust debate ahead of the law’s passing.

Of 163 orders issued so far under the Act, over a quarter were related to the Covid-19 pandemic. The most orders issued in a year – 65 – was in 2020, when the pandemic started.

Much was at stake then, and the law allowed the Government to nip falsehoods in the bud quickly, experts told ST.

Pofma was passed in Parliament on May 8, 2019, and assented to by then President Halimah Yacob on June 3 that year.

It had been a long road to get to that point.

A select committee of 10 MPs was convened in 2018 to look into the problem of deliberate online falsehoods and recommend strategies to deal with them, after Law and Home Affairs Minister K. Shanmugam floated the idea of a fake news law the year before.

The committee held public hearings over eight days in March that year, where academics, social media and mainstream media representatives, activists and others put forth their concerns. It also received 169 written representations from lawyers, journalists, businessmen, students and more, with the process culminating in a 279-page report.

Since then, Pofma has become so well known among Singaporeans that the law has become a verb. Then Prime Minister Lee Hsien Loong referenced it as such at a conference in 2023, when he said: “We have a phrase called Pofma. It has become a verb, I can Pofma you.” And Facebook users sometimes joke to one another about being careful not to “kena Pofma”.

ST’s data set showed that Pofma was mostly invoked to deal with matters related to public health and public order, as well as trust in government functions.

Dr Carol Soon, principal research fellow at the Institute of Policy Studies, who was the first to present evidence at the 2018 hearings, said it was unsurprising that most of the directions issued fell into these two of the six clauses set out in Pofma legislation: those that are “prejudicial to public health, public safety, public tranquillity or public finances” and those that “diminish public confidence in the performance of any duty or function of, or in the exercise of any power by, the Government... or a statutory board”.

The other clauses relate to security, friendly relations with other countries, election influence and inciting enmity between different groups.

She also pointed out that most corrections in the former category were issued during the pandemic, with the aim of mitigating unnecessary panic and helping to sustain trust in the Government.

“Based on the sentiments shared online, there was support for the use of Pofma in those instances,” she said.

Pofma corrections in the second category, however, demonstrate the Government’s “consistent little tolerance for actions that potentially erode people’s trust in the Government’s integrity”.

Dr Soon noted that despite false statements of fact existing in many corners of cyberspace, actions are generally taken against prominent political figures and activists.

This indicates that the Government recognises the influence these opinion leaders wield in the public sphere, she said.

The prevalence of these two categories is also reflected when looking at which ministries were the source of the most Pofma orders – the Ministry of Communications and Information (MCI) has 24, the Ministry of Health (MOH) has 23, while the Ministry of Home Affairs (MHA) has 22.

It is worth noting that MCI’s number is high because its minister is the one who issues directives for Declared Online Locations, disabling orders and access blocking orders, while the MOH and MHA numbers comprise various correction directions.

Among the ministers, Mr Shanmugam has issued the most Pofma directives – 28 – followed by Minister for Communications and Information Josephine Teo with 21 and Minister for Health Ong Ye Kung with 18.

The most common topic that Pofma orders were issued in relation to was Covid-19 with 44, followed by the Ridout Road saga in 2023 with 11, and 10 involving the 10 million population issue, which received attention during the 2020 General Election.

After 2020, the year with the next highest number of Pofma-related directives was 2023, with 43 instances that were due in part to orders related to Ridout Road and a large number of Declared Online Locations.

2020 was also the first time Pofma was used during an election, during which the alternate authority for the respective ministers – in this case, the permanent secretaries – issued the Pofma orders. There were 20 correction directions issued between the day the Writ of Election was issued and Polling Day.

While the use of Pofma in these cases gives a hint as to what the Government sees as critical issues that have to be addressed quickly, not all Pofma orders are issued immediately after the offending posts are made.

The data showed that the average period between a post being made and it receiving a Pofma order was about 6.25 days, though one day was the mode – or most common time period.

Much of the fake news related to Covid-19 – such as a false claim that Covid-19 had led to Woodlands MRT station being closed, or an alleged government cover-up of daily Covid-19 case numbers – was among those speedily refuted in a day or less.

The longest period to issue an order, barring certain outliers, was 40 days. This involved a blog post about Covid-19 vaccines by local author Cheah Kit Sun on Oct 20, 2021, that opposition politician Goh Meng Seng shared on the same day.

The duo were issued Pofma directives on Nov 29, 2021.

Associate Professor Chong Ja Ian from the National University of Singapore’s department of political science said the average of 6.25 days seems to be in some tension with the original promise of speedy responses.

“The lag in response time at a time when AI is rapidly enhancing the reach, spread and modification of mis- and dis-information calls attention to whether the response time for the usage of Pofma can keep up with the pace of technological change and the complications it brings,” he said.

The Pofma Office did not respond in time to ST’s queries, including how the appeals process is considered, how the law has regulated the fake news landscape here, and about those who incur Pofma directions repeatedly.

Worries about Pofma

During the 2018 select committee hearings, the concerns raised included difficulties in defining online falsehoods and whether such measures would impinge on free speech.

Others said social media firms needed to play a more proactive role in detecting and removing falsehoods for the law to be effective, and they believed this to be unlikely to happen. One expert even alleged a “pattern of denial and inaction” from these companies in tackling inappropriate content on their platforms.

Singapore Management University law don Eugene Tan said some concerns about Pofma – such as the law being a curb on freedom of speech, its use to deal with political dissent, and the risk of its being abused – will never go away.

“These legitimate concerns will have to be balanced against the reality that online falsehoods and manipulation have grown significantly since the select committee hearings,” he said.

It is no longer a question of why Singapore should have a law like Pofma, but how it can reap the benefits of Pofma while mitigating its downsides, he added.

Pofma becomes another tool in Singapore’s arsenal to better deal with truth decay, alongside laws such as the Foreign Interference (Countermeasures) Act passed in October 2021 and those tackling online harms, said Professor Tan.

He added that it would seem that where the authorities are concerned, Pofma – in particular the correction direction – provides the legal institutionalisation of the “right of reply”​.

Associate Professor Alton Chua from Nanyang Technological University’s Wee Kim Wee School of Communication and Information said that when Pofma was first unveiled, there was palpable concern among academics about the law dampening public discourse and curtailing freedom of speech.

But the Government then gave the assurance that Pofma would not target the academic community, and that the law excluded opinion, criticisms, satire and parody, he noted.

“To the Government’s credit, the use of Pofma has been calibrated,” he said, though he added that it was hard to tell if academics had been “spared” or if it was due to a chilling effect of the law that caused academics to self-censor.

Prof Chong, who was among those who called for independent fact-checking platforms during the public hearings, told ST there still remains the question of potential abuse of Pofma by individuals holding political authority.

“To be clear, I am not necessarily talking about today, but citizens should be aware that political systems may be gamed by those with power. That could happen some day in Singapore, and to the degree that we have more safeguards put upfront we can reduce some of that future risk,” he said.

On the social media front, ST found that all online platforms that had been issued Pofma orders over the past five years had complied.

This included putting up notices to tell users that particular posts contained false information, blocking access in Singapore to certain pages, or alerting users that they had previously viewed clips that contained false information.

The platform that saw the most Pofma orders – whether the posts were made on that platform or if it was issued a targeted correction direction, among others – was Facebook with 110.

This was followed by individual websites at 44, X (formerly known as Twitter) with 18, Instagram with 10, TikTok with nine, LinkedIn with eight, HardwareZone with four, YouTube with three and Spotify with one.

While it complied, Facebook had at several points made known its grievances about the law.

For example, in February 2020, it was ordered to disable access to the States Times Review Facebook page for Singapore users.

A spokeswoman then criticised the order for being disproportionate and flagged concerns of the law’s potential overreach.

Facebook did not respond to ST’s queries.

Five years of data on Pofma and its application has shown that the authorities do not always go straight to asking social media firms to take action.

From what is publicly known about the Pofma process based on news releases issued by the authorities, the usual process begins with a correction direction issued to an individual or entity to tack on a correction notice to the post.

Typically, if this is not done within a day, a targeted correction direction is issued to the social media platform to tack on the notice of falsehood.

For a post made on a website, an access blocking order may be issued by the Infocomm Media Development Authority for internet service providers to block it directly if the correction direction is not heeded.

If the same entity has carried at least three different false statements of fact within six months, the minister for communications and information may then declare the platform to be a Declared Online Location.

This would mean the entity has to declare its new status publicly on the platform. It would also have to de-monetise its operations, as Pofma aims to prevent profiting from the communication of falsehoods.

The next step would be to issue a disabling order if the entity did not cooperate with declaring its Declared Online Location status. This would mean the social media platform has to restrict access to the page for Singapore users.

Of all the orders issued, 113 were correction directions, 20 were targeted correction directions and five were general correction directions. While a targeted correction direction tacks on a notice to an offending post, a general correction direction may be complied to by showing a prompt to all users on a newsfeed or through a notification.

There were four disabling orders, three access blocking orders and 16 Declared Online Locations.

However, it appears that Pofma is not necessarily the first line of defence.

In an October 2021 case involving activist Jolovan Wham, Mr Shanmugam issued a correction direction to Mr Wham for misrepresenting on Twitter what the minister had said in Parliament earlier that month.

In a statement that day, MHA said it had earlier issued letters to eight individuals and a Facebook page that published false posts misrepresenting Mr Shanmugam. The ministry said then that eight of the nine had since apologised and/or corrected their posts, apart from Mr Wham who was issued the correction direction.

Prof Tan said this case shows that the authorities are also looking at engaging those who publish falsehoods and appealing to their better angels to do what is right in respect of their misinformation or disinformation.

Even so, it appears some are not deterred.

Mr Alex Tan Zhi Xiang, who received Pofma orders directly and for five Facebook pages managed by him, such as the States Times Review, was the most frequent recipient of Pofma directives – with 27.

He is followed by alternative news site The Online Citizen, whose team later transitioned to Gutzy Asia when TOC was marked as a Declared Online Location – they had 21 in total.

The third-most frequent recipient was Reform Party chief Kenneth Jeyaretnam with 12.

Compiling the Pofma data also revealed different reactions from opposition politicians.

Mr Jeyaretnam complied with all directives – including that of declaring his five platforms as Declared Online Locations – and carried on posting.

The Singapore Democratic Party (SDP) and its leader Chee Soon Juan have also had brushes with Pofma.

While they complied, they made clear publicly that they disagreed, and in several cases made appeals and went to court.

In one case that concluded in 2021, the Court of Appeal allowed the SDP’s appeal in part. It was a landmark decision for the court as it was the first-ever opportunity for it to consider the interpretation and application of Pofma since the law’s enactment.

The apex court established the constitutionality of Pofma, and also resolved conflicting High Court conclusions as to which party bore the burden of proof – the statement maker or the minister on whose authority the Pofma direction was issued.

Ultimately, the court concluded that it lay with the statement maker.

Another opposition politician who received a Pofma directive was Mr Leong Mun Wai of the Progress Singapore Party.

The Non-Constituency MP removed his post and in February 2024 stepped down as the party’s secretary-general to “take responsibility” for his Pofma order.

When asked about these different responses to Pofma, Prof Chong said it would depend on the information involved and the preferences of the individuals and entities.

“The larger point is that if Pofma is not seen to be evenly and impartially applied, with sufficient evidence for citizens to make independent assessments of each instance of Pofma application, cynicism surrounding the law could arise, reducing its effectiveness over time,” he said.

Prof Tan said a person who is serially Pofma-ed will have a credibility deficit and suffer reputational harm.

However, one challenge is that some may expect a statement is probably not false if it is not Pofma-ed. 

The authorities will have to be judicious in the use of Pofma and use other means to deal with falsehoods, saving Pofma for the more egregious cases, said Prof Tan.

Concerns going forward

While Pofma directives have been issued to identifiable individuals and entities that have put forth misinformation and disinformation, Prof Chong said he was concerned about falsehoods that come from unknown origins.

For instance, some anonymous messages circulated during the Covid-19 pandemic alleged that the hygiene of particular races and ethnic groups contributed to the spread of the virus, propagated conspiracy theories about its origins, or listed folk remedies that could prove harmful.

“Despite the anxieties these narratives caused, it appeared that not much could be done about them – especially since there was no one to issue a directive against,” Prof Chong said.

Other growing concerns about Pofma’s effectiveness centre on the rise of artificial intelligence, with generative AI tools becoming easily accessible to the public, and with this, the potential for deepfakes to spread rapidly.

Dr Soon said that as the complexity and entanglement between truth and falsity increase, Pofma’s utility is likely to cease.

She cited her 2020 study on media and internet use during the 2020 General Election, which found that only 39 per cent of about 2,000 people surveyed were confident that regulations like Pofma reduced the amount of false information online during the election. In contrast, 41 per cent neither agreed nor disagreed, while the remaining 20 per cent disagreed.

“It suggests that members of the public understand the limitations of regulation, and that was before generative AI tools were made accessible to the masses,” she noted.

She pointed to the recent Indonesian election, where an AI-generated deepfake of late president Suharto endorsing particular candidates gained traction.

“When is a deepfake permissible and when is a deepfake not permissible? When does a deepfake cross the boundary from being creative to being unethical or deceptive? These questions require tackling issues relating to provenance and ethics of information production and dissemination, beyond what existing laws like Pofma address,” said Dr Soon.

Prof Chong said AI-generated visual, video and audio deepfakes, alongside algorithms meant to exponentially increase the reach of particular posts, can hasten the spread of information, including misinformation and disinformation.

Pofma may not work fast or precisely enough to address some of these instances.

“Going after entities and individuals or takedown orders may not be enough and may not be enough to replace transparency and education,” he said.

In today’s highly dynamic information environment, the risks of misinformation and disinformation are only likely to become more complex as conflicts and major power competition intensify, said Prof Chong.

“There may need to be multiple sources of independent fact-checking, greater data transparency to allow for pre-bunking – to pre-emptively debunk disinformation – and more extensive media literacy and civic education to better protect Singapore. These are approaches that Finland and the Baltics have tried with notable success even as they face a hostile information environment,” he added. 

Prof Chua said correction directions are premised on the rationality of readers.

“The idea of juxtaposing a government statement against an offending article is so that readers can sensibly reject the alleged falsehood and embrace the truth. In reality, deep-seated beliefs and subjective sentiments cannot be shifted easily,” he said, adding that this is worsened by AI and deepfakes that make falsehoods more realistic and convincing.

At the end of the day, Pofma cannot be the only solution to combat falsehoods, he said.

“A holistic approach calls for the involvement of different stakeholders including civil society, fact-checkers, journalists and academic institutions. It is in every­one’s interest, not just the Government’s, to develop national resilience against falsehood.”

How others tackle fake news

European Union

The European Union has the Digital Services Act (DSA), a major law designed to regulate online platforms, including social media platforms like Facebook, Twitter and TikTok that are used to disseminate disinformation. It came into force in August 2023.

It also seeks to prevent illegal and harmful activities online.

Companies that fail to comply with the DSA rules could face fines of up to 6 per cent of their global turnover and could have their services suspended in the EU as a last resort.

China

In 2016, the Chinese government criminalised the creating or spreading of rumours that undermine economic and social order in a cyber-security law.

A year later, a law required social media platforms to republish and link to news articles only from registered news media outlets. Social media platforms were prohibited from publishing their own independent articles or disseminating information without attribution.

In 2018, the authorities required microblogging sites to highlight and rebut rumours on their platforms.

United Kingdom

The Online Safety Act was first conceived in 2017, and came into force in 2023 after multiple amendments. Many of its protections are designed to hold technology companies responsible for content posted on social media.

If these firms do not remove illegal content, including sexual abuse and terrorism material and death threats, they can be fined up to 10 per cent of annual revenue or £18 million (S$31 million), whichever is higher.

Social media platforms

In a May 2021 statement on its website, Meta Platforms said it was launching new ways to inform people if they were interacting with Facebook content that had been rated by a fact-checker.

For example, if a user were to “like” a page that has repeatedly shared content that fact-checkers have rated, a pop-up would appear.

Meta has also published regular articles on its site about how the firm is preparing for various elections, such as India’s general election held recently. It covered issues such as the misuse of generative AI and measures on WhatsApp to limit people’s ability to forward messages.

From 2024, advertisers around the world are also required to disclose when they use AI or digital methods to create or alter a political or social issue advertisement in certain cases, such as if the ad contains a photorealistic image or video that depicts a real person saying or doing something he did not say or do.

Social media platforms like Meta and TikTok have also announced partnerships with third-party fact-checking networks and organisations that do not moderate content directly, but identify, review and rate content.

TikTok has also prohibited paid promotion of political content or fund raising by politicians and political parties on the platform.

Since 2023, the firm has required its community to label AI-generated content (AIGC) or edited media that shows realistic-appearing scenes or people, such as by using an AIGC label or adding a clear caption, watermark or sticker.

Source: Straits Times © SPH Media Limited. Permission required for reproduction.

MAS, CAD probing Seatrium over potential offences relating to Brazil corruption case ADV: SUSS School of Law – Apply by 15 July