In January 2026, Canadian and international reporting documented a wave of non-consensual sexualized deepfakes of identifiable women generated with X’s Grok AI. Global News also reported AI-generated child sexual abuse material on the platform. Canada’s Privacy Commissioner opened and then expanded an investigation into X. Minister Solomon pointed to Bill C-16, which was still before Parliament at the time. It received Royal Assent on June 18, 2026 as the Protecting Victims Act. The Criminal Code amendments discussed here took effect on July 18, 2026. Researchers examined 187,778 image-containing posts collected around Canada’s April 28, 2025 federal election. The detector flagged about 5.9% of the sampled images as synthetic. Posts classified as politically harmful deepfakes accounted for 0.12% of the X post views tallied in the study. X can count repeat views, and the paper does not measure how many Canadian voters saw these posts or whether any changed their minds. The rules on non-consensual intimate images and federal electoral impersonation are separate questions. > GOBLIN FACTS — start with the denominator. The image figure comes from selected posts on three platforms. The view figure comes from post counters on X. Neither tells you what share of voters saw a deepfake. Start with the Canadian election study and the cases it could see. Then turn to the law. An intimate image, a fraudulent endorsement and an impersonation during an election can all be made with similar tools and raise very different legal questions. The Protecting Victims Act, Québec’s civil remedies and the election rules each give you a different place to look. The international examples and the likeness proposals from Chapter 11 help show where those protections reach and where a person still has to ask what remedy is available. ---
Bill C-16 — what it covers
Lire cette section en français
Bill C-16 became the Protecting Victims Act on June 18, 2026. The intimate-image amendments discussed here came into force on July 18, 2026. The Act also addresses other matters, including coercive control and the circumstances in which murder is classified as first-degree murder. The rules most relevant to deepfakes are in sections 162.1 and 164.1 of the Criminal Code.
Here is what the enacted law changes:
Which synthetic images qualify. Section 162.1 covers certain electronically or mechanically made depictions, including AI-generated ones. They must show an identifiable person and be realistic enough to be mistaken for a recording of that person. The offence concerns knowingly sharing or making an intimate image available, including selling or advertising it, while knowing the person did not consent to that conduct or disregarding whether they did. Making an image by itself is not the conduct prohibited by this section.
Near nudity is included. Both real recordings and synthetic depictions can qualify if they show nudity, near nudity, exposed sexual organs or explicit sexual activity. For real recordings, the person must have had a reasonable expectation of privacy when recorded and still have it when the offence occurs. For synthetic depictions, the test is identifiability and whether the depiction could be mistaken for a real recording.
Maximum penalties. When prosecuted by indictment, the ordinary maximum for non-consensual distribution rises from five to 10 years. A 14-year maximum applies if the accused knew or should have known that the person depicted was undergoing, or had just undergone, an aggravated sexual assault when the image was made. That higher maximum does not apply to every synthetic image. The distribution and threat offences can also be prosecuted by summary conviction.
Threatening to share an intimate image. The new offence can apply even if no image is shared. It requires a knowing threat made to intimidate or be taken seriously, with knowledge that the person would not consent to the threatened conduct or recklessness about their consent. The maximum is 10 years when prosecuted by indictment. Extortion is a separate offence.
Recruiting a young person. Section 169.1 is a separate, broader offence. It can apply when someone in a position of trust, power or authority recruits or encourages a person under 18 to take part in a Criminal Code or Controlled Drugs and Substances Act offence, and the young person then takes part in that offence or a related one. The recruiter must know the young person is under 18 or be reckless about their age. An adult is deemed to be in such a position for this provision.
When the 48-hour clock starts. Under subsection 164.1(5.1), if the court finds on a balance of probabilities that the material is an intimate image in relation to which a section 162.1 offence was committed, it must order the custodian of the computer system to delete it within 48 hours after the order is made. A report to a platform does not start that clock. Section 164.1 concerns material made available through a computer system within the court’s jurisdiction. Subsection 164.1(9) also provides that orders under subsections (5) to (7) take effect only after the time for final appeal has expired. A deletion order and a criminal prosecution serve different purposes, and the amendments do not create a standalone civil damages claim.
What the intimate-image amendments do not cover. The definition turns on what the image shows and, for synthetic depictions, whether it could pass for a recording of the person. These amendments do not create a dedicated remedy for the following uses when they fall outside that definition:
- Political deepfakes (fake images of politicians, electoral interference content)
- Defamatory deepfakes that are not sexualized (fake quotes, fake speeches, fake actions)
- Commercial deepfakes (unauthorized commercial use of likeness, fake endorsements)
- Identity-fraud deepfakes (deepfake-assisted scams targeting individuals)
- Generic non-consensual deepfakes that don't fall within the specific Section 162.1 scope
The Act expands an existing criminal provision and its related removal procedure. It leaves other uses of synthetic media to the laws that already apply to them, if any. The political, commercial and fraud examples below illustrate why identifying the kind of harm matters when looking for a remedy.
The Online Harms Act backstory. The previous Parliament’s proposed Online Harms Act (Bill C-63) did not become law. Its proposed regulator and platform duties were broader. The Protecting Victims Act takes a narrower approach to intimate images through the Criminal Code and court-ordered deletion.
Official texts: Protecting Victims Act · Criminal Code, section 162.1 · Criminal Code, section 164.1
The 2025 Canadian election empirical study
Lire cette section en français
Source: Results of the 2025 Canadian election deepfake study
Livernoche and colleagues’ study appeared in the proceedings of The ACM Web Conference 2026. The authors’ accessible preprint describes posts collected April 24–29, 2025, using election-related queries on X, Bluesky and Reddit. On X, they kept posts containing an image and excluded reposts and accounts with fewer than 100 followers. The study used a detector trained on OpenFake and other datasets and used models to infer account leanings and likely intent. Those choices define what the numbers cover.
Here is what can be checked in the authors’ public preprint. The conference publication is listed separately in receipt #32.
Prevalence. Table 1 of the preprint lists 11,005 posts flagged as containing synthetic imagery out of 187,778 collected, or about 5.86%. The detector can miss synthetic images. The unflagged share is a detector result, and the study cannot give a rate for every election image.
Inferred political leaning. Adding the counts in Table 1 across the three platforms gives 9.24% of posts from accounts classified as right-leaning (5,075 of 54,901) and 3.87% from accounts classified as left-leaning (2,488 of 64,271). Those rates match the preprint’s Results section. Its abstract gives 8.66% and 4.42% without reconciling the difference. The direction agrees across the two passages. The size of the gap remains uncertain, and the labels describe accounts in the sample.
Content. The preprint’s Results section puts the benign and non-political categories together at 49.41% of detected cases, about half. The researchers inferred intent from the content. Those labels cannot tell us what the author privately meant or what a viewer believed.
Views on X. The researchers compared X’s post counters six months after posting. In the X posts they analyzed, 0.52% of the tallied views went to posts flagged as deepfakes and 0.12% went to posts classified as politically harmful deepfakes. The denominator is the views tallied for the sampled X posts. X can count repeat views, including views by the author. A post view does not show whether someone examined the image, lived in Canada, was eligible to vote or changed their vote. The comparison also cannot establish what role platform filtering played.
The qualitative finding worth foregrounding. The Canadian Digital Media Research Network (CDMRN) and the Atlantic Council’s Digital Forensic Research Lab (DFRLab) identified seven deepfakes of Mark Carney during the campaign. The videos mimicked CBC or CTV interviews and directed viewers to scam sites. These examples use political celebrity to lend credibility to a financial scam. They fall outside the intimate-image amendments, but fraud law may still apply.
A separate finding about Meta. DFRLab documented the news block on Meta during the campaign and traced scam content that still circulated there. That evidence tells us something about what could circulate on Facebook and Instagram. The Livernoche study’s view comparison is about sampled posts on X. It does not measure the effect of Meta’s news block.
The government-side perspective. The Security and Intelligence Threats to Elections (SITE) Task Force's Nathalie Drouin (National Security and Intelligence Advisor) and David Morrison (Deputy Minister of Foreign Affairs) testified publicly after the election that government officials "had expected attempts to use AI to interfere in the last election, but did not detect [significant interference]." The testimony also flagged warnings about the next election facing AI-assisted interference at substantially higher scale. The intelligence-community framing: the 2025 election was not the deepfake catastrophe some predicted, but the absence of catastrophe was partly luck and partly the maturity gap of current AI tools, and the next election cycle will likely face more sophisticated deployment.
A case beyond the ballot. The election was not the only Canadian information fight AI tools were brought to. In 2026, Cipher AI — an Edmonton-and-Regina startup spun out of the University of Regina's Centre for Artificial Intelligence, Data, and Conflict — used AI to track foreign disinformation amplifying the Alberta-separatism debate, attributing campaigns to Russian networks and to US influencers with large followings. The finding worth holding onto is the one Cipher itself foregrounded: the foreign activity amplified the separatist moment but did not manufacture it. That distinction is this chapter's whole discipline applied to a live domestic case. "Foreign bots are behind it" is the doomsday framing; "foreign actors are exploiting a real and homegrown grievance" is what the evidence supported. The same caution cuts the other way, too: the AI-tracking capability that surfaces foreign amplification can also be misused to wave away authentic domestic dissent as someone else's psyop, which is why the amplify-versus-manufacture line has to be stated as plainly as the threat.
What the evidence supports. Synthetic images showed up in the sampled election posts. The authors’ labels and X counters tell us how often their detector flagged images and how views were distributed in the sample. A separate investigation documented scams and the Meta news block. To move from those findings to a claim about election impact, ask who actually saw the content, what they took from it and what they would have done otherwise. The study leaves those questions open.
ALIGNMENT — how common before how scary. One convincing deepfake makes a great headline and a terrible measurement. Before accepting that something is "everywhere," ask for the base rate: out of how many, over what stretch of time, reaching whom? Fear scales with the vivid example. Reality scales with the denominator.
🧌 GOBLIN CHECK — a personal one. The About the Author section promised this chapter would return to Strange Harvest, so here it is, receipts and all. Our film's festival cut included roughly thirty seconds of licensed, disclosed AI-generated stills used as placeholders during the festival run. The backlash was immediate, so we pulled the material for the theatrical release. Here's the part that belongs in this chapter: the accusations didn't stop — they migrated. Commenters stayed certain, now pointing at sequences that were entirely practical, or at compositing that was merely budget-priced. The lesson the study above documents at scale, I got to live at retail: detection panic doesn't track what's actually synthetic. A label, once applied, doesn't come off. And the audience most confident it could "always tell" performed, on my film, no better than a coin flip. So when you read the 5.9% prevalence number, hold the other side of the coin too: a public that can't reliably spot the 5.9% also can't reliably clear the other 94.1%. The harm isn't only fake content believed. It's real work disbelieved. — A.Y.
The political deepfake regulatory gap
Lire cette section en français
Bill C-25 received royal assent on June 18, 2026. It expanded the federal electoral impersonation offence in section 480.1 of the Canada Elections Act. With intent to mislead, creating or commissioning specified deceptive images, voice imitations or false audio recordings of listed people can be an offence. So can distributing, transmitting or publishing those materials with that intent. The list includes candidates, certain party leaders and election officials. The provision makes an exception where the person establishes that the material was manifestly for parody or satire. This is separate from the Criminal Code’s intimate-image rules under Bill C-16.
The Protecting Victims Act expands the Criminal Code’s intimate-image provisions. It does not add a dedicated offence for deceptive election images or videos. In those cases, any remedy depends on the facts and on other laws, including election, defamation and fraud law.
EXAMPLE — the smoke detector wired to one room. The new intimate-image rules cover the kitchen. A fabricated campaign speech, endorsement or scam video is in another room. For those, start with what the clip was used to do: mislead voters, damage a reputation or take someone’s money. That helps identify which other laws might apply.
What existing Canadian law might apply to political deepfakes, with the gaps:
Defamation. Canadian defamation law could in principle apply to deepfakes that depict identifiable individuals saying or doing things that damage reputation. The challenges: defamation litigation is slow, expensive, requires the victim to bring civil action, and produces remedies (damages) that don't reach the underlying viral content. Anonymous or pseudonymous creators are difficult to identify. The platform hosting the content has limited liability under section 230-equivalent frameworks. Defamation is a real but limited remedy.
The Canada Elections Act. Section 480.1 turns on the person depicted, the conduct and the intent to mislead. A fabricated endorsement involving someone outside the statutory list calls for a separate analysis under other applicable laws.
Privacy law. Provincial privacy laws and the federal PIPEDA could in principle apply to non-consensual use of a person's image, but the application to synthetic-image cases is unsettled and varies by jurisdiction.
Criminal Code provisions other than Section 162.1. Identity theft, fraud, criminal harassment, and uttering threats provisions could apply in specific cases but are not designed for the bulk-deployment scenarios deepfake-generation enables.
Common-law tort. Various torts (misappropriation of personality, intrusion upon seclusion, intentional infliction of emotional distress) could apply but require costly litigation and produce limited remedies.
What's missing from the Canadian framework that peer jurisdictions have:
- A specific criminal prohibition on non-consensual deepfakes generally (not limited to sexual content). This exists in Australia, the UK, and various US states.
- Removal and civil remedies. Canada’s 48-hour clock starts when a court makes the order. In the US, it starts when a covered platform receives a valid TAKE IT DOWN Act request. The Canadian amendments do not create a separate federal civil damages regime.
- Significant financial penalties for non-compliance that create real platform-side incentives. The UK Online Safety Act allows fines up to £18 million or 10% of global annual revenue. Australia's framework includes civil penalties up to AU$782,500 per violation. Section 480.1 creates an impersonation offence and sets no platform-wide compliance duties.
- Mandatory disclosure of AI-generated content in certain contexts. The EU AI Act requires disclosure that content is AI-generated in many use cases. The US has state-level requirements (California, Texas) for political AI disclosure. Canada has no general AI content disclosure requirement.
- Electoral disclosure. Some US states use election-specific restrictions or disclosure rules, including California's contested AB 2839. Canada’s section 480.1 targets specified impersonation and deceptive media involving listed people. It imposes no general labelling duty for synthetic campaign content.
The structural picture. Federal law now has targeted rules for qualifying intimate images and for specified electoral impersonation. For other uses, the available remedies depend on the facts and the law that applies.
A specific finding that connects to the Chapter 5 leaked draft analysis: the CBC obtained version of AI for All contained more detailed engagement with deepfake-related concerns than the final announcement. The narrowing between draft and announcement reflects the same pattern visible across the strategy: internal candour about challenges, external presentation focused on commitments. The deepfake regulatory gap is one of the specific cases where the gap between internal recognition and external commitment is observable.
Quebec leadership returning
Lire cette section en français
A pattern this guide has named in Chapter 5, Chapter 9, and Chapter 10 surfaces again: on AI-adjacent policy areas where the federal government has been silent or slow, Quebec has consistently moved earlier and more comprehensively. The pattern holds for deepfakes specifically.
Quebec's Act respecting the rights of action against persons distributing intimate images (officially in force in Quebec in 2024) establishes a civil regime for non-consensual sharing of intimate images, including AI-generated synthetic images, that is distinct from and complementary to federal criminal law. The Quebec regime provides:
- A civil cause of action allowing victims to seek remedies without requiring criminal prosecution
- Statutory damages that can be awarded without proving specific harm, providing a remedy in cases where actual damages are difficult to quantify
- Penalties of up to CA$5,000 per day for individuals and CA$50,000 per day for legal persons (companies, platforms) that fail to comply with takedown orders
- A faster procedural pathway than ordinary civil litigation, designed to address the time-sensitivity of online content
- Specific provisions for platform liability that go beyond what federal law provides
The Quebec framework is not perfect. Implementation has been gradual, awareness among potential users is limited, and the regime's scope is narrower than what fully comprehensive deepfake regulation would address. But it represents the most significant Canadian deepfake response currently in force for the specific category of non-consensual intimate images.
Quebec has a separate civil process for intimate images. The federal Criminal Code provisions apply across Canada, including in Quebec. They do not replace Quebec’s process or extend that provincial process to the rest of the country. The federal deletion deadline starts with a court order.
Whether other jurisdictions should adopt a comparable civil process is a policy question. It is separate from what the federal Criminal Code amendments now provide.
The international comparison framework
Lire cette section en français
Putting Canada's position in the comparative context provides the structural picture readers need to evaluate the regulatory choices:
United States. Under the TAKE IT DOWN Act, covered platforms must provide a way to request removal of non-consensual intimate images, including qualifying AI-generated ones, and must act within 48 hours of a valid request. The Federal Trade Commission enforces that process. The trigger differs from Canada’s court order. State laws vary and may address other uses of deepfakes.
United Kingdom. The Online Safety Act (in force 2023, with implementation phasing through 2025–2026) establishes comprehensive online-harms regulation including deepfake provisions. Ofcom is the regulator. Fines up to £18 million or 10% of global annual revenue. Specific provisions for non-consensual intimate images, harm to children, and various other categories. The Act has been controversial in implementation (free-speech concerns, encryption disputes) but represents the most comprehensive single-jurisdiction online-harms framework currently in force in a Westminster-system democracy.
Australia. The eSafety Commissioner has specific authority over online safety including deepfakes. The framework includes 2024 amendments specifically addressing AI-generated content. Civil penalties up to AU$782,500 per violation. Criminal penalties of up to six years' imprisonment (seven in aggravated cases) for sharing non-consensual sexually explicit deepfakes, under the 2024 federal amendments. Comprehensive scope including children's safety, image-based abuse, and harassment.
European Union. The AI Act (effective August 2024, with phased implementation) establishes risk-based AI regulation including transparency requirements for AI-generated content. The Digital Services Act (effective 2024) establishes notice-and-action obligations for online platforms. Combined, these provide the most comprehensive regulatory framework for AI-generated content globally. Significant fines (up to 7% of global annual turnover under the AI Act for the most serious violations; up to 6% of worldwide turnover under the DSA).
Singapore. Substantial framework including the Protection from Online Falsehoods and Manipulation Act (POFMA) and complementary legislation specifically addressing AI-generated content.
South Korea. Specific legal framework including criminal provisions for deepfakes with substantial penalties.
Japan. Active regulatory engagement including AI-content-disclosure proposals.
Canada. The Protecting Victims Act extends federal criminal law to certain realistic synthetic intimate images and requires deletion within 48 hours after a qualifying court order. The US deadline instead runs from a valid request received by a covered platform. Quebec also has a separate civil process for intimate images. The federal amendments do not create a dedicated political-deepfake offence, a general AI disclosure rule or a platform regulator. Separately, section 480.1 of the Canada Elections Act covers specified deceptive images and audio involving listed people.
The comparison shows where the tools differ. Canada has targeted electoral and intimate-image offences. Broader platform duties, civil remedies and AI-content disclosure remain distinct policy questions.
The convergent likeness rights mechanism
Lire cette section en français
Chapter 11 introduced a specific finding: ACTRA's NIL Rights argument (creator-rights framing) and CIGI's likeness-rights-in-copyright argument (privacy-and-harm-prevention framing) converge on the same legislative mechanism — the recognized likeness right defined in Chapter 11. This convergent mechanism is the most concrete policy path the guide has identified that engages both the creator-rights debate from Chapter 11 and the deepfake regulation gap from this chapter.
Working through how a likeness rights framework would address the gaps:
For sexualized deepfakes. The federal amendments provide criminal offences and a court deletion process for covered intimate images. The proposed likeness-rights framework would add a different option: a civil action brought by the person depicted, potentially with statutory damages. That proposal is separate from the remedies in the Protecting Victims Act.
For political deepfakes. Section 480.1 provides a federal offence where the statutory conditions are met. The proposed likeness framework would serve a different purpose by giving people a civil claim over specified unauthorized uses of their likeness, including uses outside that electoral offence.
For commercial deepfakes. The current framework relies on defamation, common-law misappropriation of personality, and contract-based protections (which ACTRA's collective agreements exemplify) — a remedy scattered across several legal mechanisms. A likeness rights framework would consolidate it into one generalized remedy applicable across the workforce and the population.
For identity-fraud deepfakes. The recent Carney scam-deepfake case illustrates a category where current law is fragmented — criminal fraud applies to the scam, but the deepfake itself sits in the gap between criminal and civil remedies. A likeness rights framework would provide the depicted individual with a direct civil cause of action against unauthorized commercial use of their likeness, complementary to whatever criminal action follows the underlying scam.
The Danish model that CIGI specifically cites. Denmark's proposed Copyright Act amendment would establish likeness rights as part of copyright law: a new section giving individuals exclusive rights over realistic digital imitations of their appearance and voice, protection running fifty years past death, with explicit carve-outs for caricature, satire, parody, and criticism of power. The amendment was put out for consultation in mid-2025, notified to the European Commission that October, and still awaits final passage in the Folketing as of mid-2026. The right to one's image, voice, and likeness joins the same legal framework that protects creative works. Enforcement uses the same mechanisms (takedown orders, statutory damages, injunctive relief) that copyright enforcement uses. The model has the advantage of building on existing infrastructure rather than requiring an entirely new regulatory regime.
The convergent mechanism's significance for Canadian policy. A single piece of Canadian legislation could close substantial portions of the Chapter 11 IP/copyright gap and the Chapter 13 deepfake gap simultaneously. It would not satisfy every position — TWUC and ACTRA would still pursue compensation frameworks beyond what likeness rights provide, civil liberties advocates would scrutinize the scope of restrictions on synthetic content, and Geist/Craig might worry about overextension into legitimate satirical and educational use. But it would advance both debates meaningfully through a single legislative vehicle. The federal AI strategy could include this commitment. AI for All does not currently include it. Chapter 20 returns to the question of how it might.
The working test
Lire cette section en français
Ask what kind of deepfake it is, what laws cover the conduct and which platforms, dates and posts the evidence covers. Start with the denominator. The study’s 0.12% of the sampled X post views comes from counters read six months later. Who was exposed is a separate question. Can you reproduce the rates from the table? Check how the authors inferred political leaning and who was left out.
What the evidence actually licenses
Lire cette section en français
CHAPTER RECAP — you now have: - The enacted Protecting Victims Act: which intimate images qualify, how it handles distribution and threats, the 10- and 14-year maximums on indictment, and why the 48-hour deletion clock starts with a court order. The youth-recruitment offence is separate, and the intimate-image rules do not create a dedicated political-deepfake offence. - What the sampled posts reveal about synthetic images and inferred account leanings, what X’s repeatable post views can measure, and why the separate evidence about scams and the Meta news block answers a different question. - The electoral impersonation rules under Bill C-25, including the listed people, covered conduct, intent to mislead and the exception for manifest parody or satire. - Québec's intimate-image civil regime includes court orders and a separate penalty for failing to comply. The daily fines can reach CA$5,000 for a natural person and CA$50,000 in other cases. - The international comparison, including differences in electoral coverage, platform duties, civil remedies and AI-content disclosure. - The convergent likeness rights mechanism — ACTRA's NIL Rights and CIGI's likeness-rights-in-copyright converging on the Danish-model framework that could close substantial portions of both the Chapter 11 IP gap and the Chapter 13 deepfake gap through a single legislative vehicle. - The working test for evaluating any deepfake-related claim: category, scope, reach, remedy, verification, politics.
Someone still has to check the clip that lands in your feed. Chapter 14 asks who pays for that work when the newsroom doing it is already stretched.
The study measures detector flags and recorded post views in a sample. The 2026 amendments change the law. Read them together to ask which conduct the evidence documents and which conduct the statutes cover.
Bias label for this chapter: empirical-and-regulatory analysis of Canadian deepfake landscape, with specific commitment to presenting the 2025 election study findings in their nuanced framing rather than collapsing them toward either alarm or dismissal. Author lean: skeptical of both "deepfakes broke the election" and "deepfakes don't matter" framings; sympathetic to the convergent likeness rights mechanism as a path through current regulatory fragmentation; willing to name the structural gap between Canadian framework and peer-country frameworks as the empirically dominant feature of the current situation; explicit about the limits of empirical evidence from a single election in projecting future deepfake-deployment patterns. Government framing (Bill C-16, ministerial statements) treated as primary on legislative commitments. Academic-empirical sources (the 2025 election study, DFRLab analysis) treated as primary on factual claims about prevalence and reach. Civil society documentation (CIGI on likeness rights, ACTRA on NIL Rights) treated as primary on policy positions. International peer-country frameworks treated as primary on legislative content.
Primary sources cited or relied on in this chapter: Protecting Victims Act (S.C. 2026, c. 19); Global News and CBC reporting on the January 2026 Grok deepfake scandal and the Office of the Privacy Commissioner's investigation of X (2026); Criminal Code sections 162.1 and 164.1; "Deepfakes in the 2025 Canadian Election: Prevalence, Partisanship, and Platform Dynamics" (ACM Web Conference 2026 / arxiv:2512.13915); Livernoche et al. (2025) OpenFake dataset documentation; Canadian Digital Media Research Network and Digital Forensic Research Lab analyses of the 2025 Canadian federal election; SITE Task Force public testimony post-election (Nathalie Drouin, David Morrison); Quebec Act respecting non-consensual sharing of intimate images (2024); Centre for International Governance Innovation analysis "Canada must do more to protect women and girls from harmful deepfakes" (December 2025); Suzie Dunn (Dalhousie University) academic work on non-consensual synthetic intimate images; US TAKE IT DOWN Act (2025); UK Online Safety Act (2023, in force 2023–2026); Australian eSafety Commissioner framework; European Union AI Act (effective August 2024); EU Digital Services Act (effective 2024); Denmark's likeness-rights Copyright Act amendment (introduced 2025). Detailed citations in the Sources appendix.
🧌 GOBLIN CHECK — a personal one. The About the Author section promised this chapter would return to Strange Harvest, so here it is, receipts and all. Our film's festival cut included roughly thirty seconds of licensed, disclosed AI-generated stills used as placeholders during the festival run. The backlash was immediate, so we pulled the material for the theatrical release. Here's the part that belongs in this chapter: the accusations didn't stop — they migrated. Commenters stayed certain, now pointing at sequences that were entirely practical, or at compositing that was merely budget-priced. The lesson the study above documents at scale, I got to live at retail: detection panic doesn't track what's actually synthetic. A label, once applied, doesn't come off. And the audience most confident it could "always tell" performed, on my film, no better than a coin flip. So when you read the 5.9% prevalence number, hold the other side of the coin too: a public that can't reliably spot the 5.9% also can't reliably clear the other 94.1%. The harm isn't only fake content believed. It's real work disbelieved. — A.Y.
Recap
- The enacted Protecting Victims Act: which intimate images qualify, how it handles distribution and threats, the 10- and 14-year maximums on indictment, and why the 48-hour deletion clock starts with a court order. The youth-recruitment offence is separate, and the intimate-image rules do not create a dedicated political-deepfake offence.
- What the sampled posts reveal about synthetic images and inferred account leanings, what X’s repeatable post views can measure, and why the separate evidence about scams and the Meta news block answers a different question.
- The electoral impersonation rules under Bill C-25, including the listed people, covered conduct, intent to mislead and the exception for manifest parody or satire.
- Québec's intimate-image civil regime includes court orders and a separate penalty for failing to comply. The daily fines can reach CA$5,000 for a natural person and CA$50,000 in other cases.
- The international comparison, including differences in electoral coverage, platform duties, civil remedies and AI-content disclosure.
- The convergent likeness rights mechanism — ACTRA's NIL Rights and CIGI's likeness-rights-in-copyright converging on the Danish-model framework that could close substantial portions of both the Chapter 11 IP gap and the Chapter 13 deepfake gap through a single legislative vehicle.
- The working test for evaluating any deepfake-related claim: category, scope, reach, remedy, verification, politics.
Sources
- Protecting Victims Act (S.C. 2026, c. 19)
- Global News and CBC reporting on the January 2026 Grok deepfake scandal and the Office of the Privacy Commissioner's investigation of X (2026)
- Criminal Code sections 162.1 and 164.1
- "Deepfakes in the 2025 Canadian Election: Prevalence, Partisanship, and Platform Dynamics" (ACM Web Conference 2026 / arxiv:2512.13915)
- Livernoche et al. (2025) OpenFake dataset documentation
- Canadian Digital Media Research Network and Digital Forensic Research Lab analyses of the 2025 Canadian federal election
- SITE Task Force public testimony post-election (Nathalie Drouin, David Morrison)
- Quebec Act respecting non-consensual sharing of intimate images (2024)
- Centre for International Governance Innovation analysis "Canada must do more to protect women and girls from harmful deepfakes" (December 2025)
- Suzie Dunn (Dalhousie University) academic work on non-consensual synthetic intimate images
- US TAKE IT DOWN Act (2025)
- UK Online Safety Act (2023, in force 2023–2026)
- Australian eSafety Commissioner framework
- European Union AI Act (effective August 2024)
- EU Digital Services Act (effective 2024)
- Denmark's likeness-rights Copyright Act amendment (introduced 2025).