Comparative Project for LAW_V 406D
An interactive comparative brief on deepfakes, scams, and more.
Audience: This website is designed primarily for law students, researchers, and legal practitioners interested in how Singapore and Canada respond to AI-enabled victimisation.
Project Scope
This website compares how Singapore and Canada respond when artificial intelligence is used to victimise individuals, especially through sexual deepfakes, intimate-image abuse, and AI-assisted impersonation scams.
The key comparison is not whether either jurisdiction has law. Both do. The more important question is what each system adds when ordinary criminal law is not enough: Canada tends to rely more heavily on existing criminal-law categories, while Singapore supplements these with more explicit online-harms and victim-relief measures.
The significance of the topic is highlighted through reports and surveys globally. One poll published in South Korea in March 2026 indicated that “More than 80 percent of Korea's teenagers and adults expressed concerns over online abuse involving the misuse of generative artificial intelligence (AI) tools” (Korea Times poll). Meanwhile, a report submitted pursuant to a UN General Assembly resolution (77/193) showed that technology-facilitated violence against women and girls has been intensifying, with studies showing 16 to 58 per cent of women worldwide impacted (UN Women report).
Main Thesis
Canada relies more heavily on technology-neutral criminal-law reasoning through existing offences. Singapore also uses penal law, but supplements it with a more explicit online-harms and victim-relief structure.
Singapore
Stronger on containing harm once harmful content or conduct is already circulating.
Canada
Stronger on offence continuity and doctrinal adaptation.
AI can generate or alter realistic sexual or compromising images, amplifying humiliation and reputational harm.
AI also intensifies fraud by making deception cheaper, faster, and more convincing through synthetic identities and voices.
The real contrast lies in whether law only punishes after the fact, or can also interrupt harm while it is spreading.
This website is organised by issue area. Each section asks three questions: what harm is involved, whether existing offences already capture it, and what remedies or accountability mechanisms exist once harm begins to spread.
Deepfake abuse, intimate-image manipulation, or AI-assisted impersonation.
Fraud, personation, intimate-image offences, or child sexual abuse material law.
Takedown, disruption, evidentiary control, oversight, and victim relief.
Issue Area One
Deepfakes and intimate-image abuse provide the clearest test of whether existing law can respond to realistic but wholly synthetic sexual representations. This is also the issue area where victim relief and takedown mechanisms matter most.
Source: Channel NewsAsia report on IMDA letters of caution to X and TikTok
Singapore does not rely only on later punishment. Its broader architecture is aimed at stopping and reducing harm while it is occurring.
Even in a stronger interventionist framework, wholly synthetic deepfakes may still raise interpretive questions. Furthermore, it may raise freedom/rights concerns.
R v Larouche, 2023 QCCQ 1853, a key Canadian case dealing with the offence of making child pornography using deepfakes, had the court describe deepfake technology as using deep-learning algorithms to create extremely realistic fakes, including face-swaps and synthetic voices, and noted that the offender had produced such content. The significance of Larouche is that it shows Canadian courts can treat AI-generated sexual exploitation of children as falling within existing child-pornography offences rather than as conduct outside the criminal law merely because the imagery is synthetic. That point is reinforced by the Supreme Court of Canada’s discussion in Quebec (Attorney General) v Senneville, 2025 SCC 33. The Court stated that child pornography is particularly harmful “whether or not it depicts real children,” and that such material “can now be generated on a very large scale using artificial intelligence,” and cited Larouche as an example. This is doctrinally important because it shows the Supreme Court expressly recognising AI-generated child sexual abuse material as a live and serious criminal-law problem, not an imagined future issue.
Analysis: British Columbia now has a serious victim-relief regime for deepfake and intimate-image harms. This means the Canada–Singapore gap is much narrower than it first appears in this specific area.
Use this to compare the same issue across both jurisdictions more directly.
Issue Area Two
In the scams context, the offence gap is smaller. Both jurisdictions already criminalise deception and personation. The main difference lies less in offence definition than in enforcement structure, speed of intervention, and remedial architecture.
Singapore’s offence structure captures scams comfortably. For scams, the Penal Code 1871 includes s 415 on cheating, s 416 on cheating by personation, and s 416B on cheating by remote communication. These provisions are broad enough to capture AI-assisted impersonation fraud, including scams using voice cloning, because the legal wrong remains deception causing harmful inducement.
This is the core Singapore scam offence. It is broad enough to capture AI-assisted deception, including voice-cloning or other synthetic impersonation, because the legal wrong remains deceit causing harmful inducement.
This provision is especially important for AI-enabled impersonation scams. A cloned voice, fake avatar, or synthetic identity can still fit within cheating by personation if it is used to deceive the victim into acting on a false belief about who is communicating with them.
OCHA is not the underlying scam offence itself. Its significance is remedial: it gives Singapore a disruption-oriented framework to interfere with online criminal activity quickly, rather than waiting only for ordinary prosecution.
Canada’s existing fraud and identity offences do much of the work. The Criminal Code already contains broad fraud and identity-related offences. Section 380(1) criminalises fraud by “deceit, falsehood or other fraudulent means”; s 403(1) criminalises fraudulently personating another person; and s 402.2(1) criminalises obtaining or possessing identity information with intent to use it to commit indictable offences involving fraud, deceit, or falsehood. Those provisions are flexible enough to capture voice-cloning scams and similar AI-enabled impersonation schemes, because the gravamen of the wrong remains dishonest deception and misuse of another’s identity. The technology changes the persuasiveness of the deception, but not the basic legal structure of the offence.
This is the central Canadian offence for AI-assisted scams. A voice-cloning scam remains fraud because the essence of the wrong is still deceit and loss, even if the deception is technologically enhanced.
This provision is useful where AI-assisted scams involve obtaining, holding, or trafficking in identity information for fraud or personation-based offences.
Issue Area Three
This includes transparency, reasons-giving, auditability, safeguards against authorities' abuse, and more.
Singapore
Stronger on takedown, disruption, service-level intervention, and redress. However, comes with questions on rights and lack of independent oversight/government overreach.
Canada
Careful not to grant government and authorities overly sweeping powers, but this means slower remedy framework.
Critics argued that the Bill gave the state and regulator very strong powers with insufficient safeguards.
First, they said the threshold of “reason to suspect” was too low. In their view, it was too subjective, could be abused in borderline disputes, and should be replaced with a stricter threshold such as “reasonable grounds to believe.”
Secondly, they argued that the appeals and oversight structure was inadequate. The concern was that the Online Safety Appeal Committee was not sufficiently independent because its members are appointed by the Minister, and that there was no meaningful merits appeal to the High Court. Judicial review was seen as too limited because it focuses on legality of process rather than whether the decision was substantively right.
Thirdly, critics raised free speech and public interest concerns. They argued that the Bill lacked clear carve-outs for legitimate expression, such as fair comment on matters of public interest or investigative journalism. Without such safeguards, they feared the law could chill lawful criticism or reporting.
Fourthly, they raised concerns about transparency and accountability. They said that such extensive powers should be matched by requirements such as annual reporting to Parliament, publication of reasons, and clearer institutional accountability. There was particular concern that the Commissioner would not always be required to give detailed reasons at the reconsideration stage.
Finally, critics pointed to ambiguities and exemptions, including the exemption for public agencies, the breadth of certain definitions, and the possibility of opaque “shadow ban” style measures through engagement-reduction directions.
Government responses
The Government’s central response was that the Bill is meant to provide fast, victim-centred relief, and that many of the proposed safeguards would undermine that purpose.
On the low evidentiary threshold, the Government said “reason to suspect” was necessary for speed. A higher threshold would require more investigation before action could be taken, leaving victims exposed for longer. It also argued that this threshold was not novel, since it was modelled on existing protective regimes such as the Criminal Procedure Code and OCHA, and that false reporting is itself a criminal offence.
On appeals and oversight, the Government argued that allowing broader court appeals would create a war of attrition, especially where powerful platforms or well-resourced parties could prolong proceedings and wear victims down. It said judicial review remained available as a safeguard, while avoiding turning each case into lengthy litigation.
On free speech concerns, the response was that explicit “fair comment” exceptions were unnecessary and potentially risky. The Government’s position was that genuinely civil and respectful public-interest speech should not satisfy the Bill’s thresholds for harmful conduct in the first place, while an express carve-out might create loopholes for abusive actors.
On transparency, the Government accepted that accountability matters, but said mandatory reporting requirements might burden a new agency too heavily at the outset. Instead, it indicated that regular public reporting could still be considered on a discretionary basis, provided it did not further harm victims.
On engagement-reduction directions and similar tools, the Government defended them as necessary to deal not only with the harmful post itself, but with the wider systemic amplification that makes online harms spread. It also presented tools such as rights of reply as more balanced remedies than simple takedown alone.
Overall takeaway
The criticism was essentially that the Bill risked overreach, insufficient independent review, and chilling of legitimate speech. The Government’s answer was that speed, practicality, and victim protection justified a more interventionist model, and that too many additional safeguards would make the system slow, expensive, and ineffective.
It appears that concerns over rights and overly broad wording has previously held back certain AI-based legislation. This includes a Bill C-63, which critics argued bundled several very different ideas into one bill, and the most controversial parts were seen as overbroad and risky for free expression. Critics argued it could chill speech, create vague or sweeping new powers, and overload the human-rights system with complaints. Bill C-63 had been described was to be "An Act to enact the Online Harms Act, to amend the Criminal Code, the Canadian Human Rights Act and An Act respecting the mandatory reporting of Internet child pornography by persons who provide an Internet service and to make consequential and related amendments to other Acts."
B.C. Civil Liberties Association commentary on Bill C-63
The same occurred with Bill C-27. The bill drew heavy criticism for being too vague, too broad, and not specific enough about how AI systems would be classified and enforced. Critics also objected to the consultation process and the lack of clear independent oversight.
Montreal AI Ethics Institute article on AIDA / Bill C-27
One author describes this as follows: "The AIDA was tabled in Canada’s House of Commons in June 2022 with the ambitious goal of establishing a comprehensive regulatory framework for AI systems across Canada. However, the AIDA was embroiled in controversy throughout its life in Parliament. A chorus of individuals and organizations voiced concern with the AIDA, citing its exclusionary public consultation process, its vague scope and requirements, and its lack of independent regulatory oversight as reasons why the legislation should not become law. Though the government ultimately proposed some amendments to the AIDA in response to criticisms, the amendments did not sufficiently address the fundamental flaws in the AIDA’s drafting and development. As a result, the AIDA languished and died in a parliamentary committee, unable to secure the confidence and political will needed to proceed through the legislative process."
Nonetheless, as discussed earlier, there still exists state-driven legislation, albeit not on a federal level.
Final Evaluation
This dashboard summarises the comparative evaluation across the main dimensions of the project.
Singapore: stronger on breadth and institutional integration, with explicit online-harms and victim-relief mechanisms extending beyond ordinary penal law.
Canada: narrower at the federal level, but stronger than it first appears in some provinces. British Columbia’s Intimate Images Protection Act significantly narrows the gap for intimate-image and deepfake-related relief.
Singapore: existing deception and personation offences already capture AI-assisted scams, with OCHA adding disruption-oriented powers.
Canada: fraud, personation, and identity-information offences already do substantial work.
Singapore: stronger on speed and victim-centred intervention, but this raises sharper concerns about overreach, oversight, and the adequacy of review mechanisms.
Canada: more cautious about broad state power and generally stronger on procedural safeguards, transparency concerns, and resistance to overbroad regulation, though this can produce a slower and more fragmented response.
Neither jurisdiction offers a complete model. Singapore currently provides the more integrated and intervention-capable response to AI-enabled victimisation, especially where speed matters. Canada, however, offers greater caution against overreach, stronger doctrinal continuity, and more robust process-based safeguards. The most persuasive approach would combine Canada’s rights-conscious legal restraint with Singapore’s faster and more centralised victim-relief architecture.
Authority Base
Statutory texts referred to in this website are drawn primarily from Singapore Statutes Online and the Canadian Justice Laws Website. Extensive credit also goes to Professor Benjamin Perrin, for his course and lessons in LAW_V 406D - Topics in Criminal Justice (Artificial Intelligence), and his book Artificial Intelligence & Criminal Justice: Cases and Commentary, 2024 CanLIIDocs 3035
AI Disclosure
Generative AI tools were used for research assistance, drafting support, and web design guidance and generation of code and images. Legal analysis, statutory checking, source selection, and final editorial decisions were by the author.