AI Safety
Who Is Affected, and How?
AI is not a single technology with a single impact. Its risks and harms land differently depending on who you are, what work you do, and what stage of life you are in. This page examines how AI is affecting Canadians across six groups, in depth, in plain language, with specific attention to Canadian law and policy.
AI affects young Canadians before they can meaningfully consent to it
Young Canadians are growing up inside AI-shaped environments: social media feeds designed to maximize engagement, school systems beginning to deploy AI monitoring tools, and a job market being restructured faster than educational institutions can respond. Unlike most technologies of the past, AI is influencing young people's development, beliefs, and opportunities at scale, with limited oversight and fewer protections than exist in other high-stakes domains.
Social media algorithms and mental health
The recommendation algorithms powering TikTok, Instagram, YouTube, and Snapchat have one primary objective: keeping users on the platform as long as possible. For young people, this creates a system that is extraordinarily powerful and largely invisible. Meta's own internal research, leaked in 2021, found that Instagram made body image issues significantly worse for teenage girls. YouTube's recommendation algorithm has been documented routing teenagers from mainstream content toward increasingly extreme material, not because anyone programmed it to, but because outrage and intensity drive engagement, and the algorithm learned this.
These are not design oversights companies simply haven't gotten around to fixing. They are the output of systems working exactly as intended, with goals that are not aligned with the wellbeing of the young people using them.
Deepfakes and image-based abuse targeting minors
AI image generation tools have created a crisis for young people, particularly young women. In 2024, students at a school in New Jersey used AI tools to create non-consensual nude images of female classmates and share them through group chats. Similar incidents have been reported across Canada. The tools required are free, require no technical skill, and produce results that are difficult to distinguish from photographs.
Creating and distributing non-consensual intimate images in Canada is a criminal offence under section 162.1 of the Criminal Code. But the law was written before AI image generation existed, enforcement capacity has not kept pace with the technology, and the harm, which spreads instantly, is often irreversible by the time any legal action is possible.
AI surveillance in schools
Schools in Canada and the United States have begun deploying AI-powered monitoring tools: systems that track student keystrokes to detect "off-task" behaviour, AI proctoring software that watches eye movements during online exams, and behavioural analytics platforms that flag students for counsellor attention based on algorithmic scores. These systems generate data about children that can follow them through school, and sometimes beyond. The accuracy of these tools, particularly across different demographic groups, is rarely audited. Consent processes are often inadequate.
The job market young Canadians are entering
Students entering the workforce today are doing so during an AI-driven disruption unlike any previous generation has faced. The entry-level roles that have historically served as career stepping-stones, including data processing, content moderation, paralegal research, copywriting, and basic design, are being automated at scale. The skills educational institutions are training students to develop may have a substantially shorter shelf life than at any previous point in modern history.
This is not a reason to avoid education. It is a reason to ask hard questions about which skills remain valuable as AI capabilities expand, and whether institutions training the next generation are adapting fast enough.
Algorithmic radicalization
Young people spend more time online than any other demographic and are more susceptible to algorithmic influence on their beliefs and identity. Documented "radicalization pipelines", pathways from mainstream content to extremist communities maintained by recommendation algorithms, disproportionately affect young men in their teens and early twenties. These patterns have been documented by researchers at universities across Canada, the US, and UK, and acknowledged, if not meaningfully acted on, by the platforms themselves.
The core issue: Young people are among the most exposed to AI systems and among the least empowered to push back against them. AI governance that does not specifically address the risks to minors, including algorithmic harm, deepfake abuse, and surveillance in schools, is governance that is leaving its most vulnerable citizens unprotected.
What to watch for in Canada
- Whether Canada's proposed AI legislation (Bill C‑27 / AIDA) includes specific protections for minors
- Whether provincial education ministries are auditing AI surveillance tools deployed in schools
- Whether the Criminal Code is updated to address AI-generated non-consensual intimate images with penalties that reflect the scale of harm
- Whether Canada requires age verification on AI image generation platforms
Women bear a disproportionate share of AI’s most immediate harms
AI's harms are not distributed equally. Women are disproportionately targeted by AI-enabled harassment and image abuse, systematically disadvantaged by AI hiring and diagnostic tools trained on biased data, and at greater risk from the labour disruptions AI is driving. At the same time, women are underrepresented in the rooms where AI policy is being made and in the technical teams building these systems.
Non-consensual intimate imagery
AI image generation has dramatically lowered the barrier to creating realistic fake intimate images of real people. Research by Sensity AI found that 96% of deepfake videos online are non-consensual intimate images, and the overwhelming majority of victims are women. This technology is being used against public figures and private individuals alike. In Canada, the creation and distribution of non-consensual intimate images is a criminal offence. But criminal law operates slowly; the images spread instantly.
The harm is not only individual. Women in public life, including politicians, journalists, academics, and activists, report that the threat of deepfake abuse creates a chilling effect on their participation in public discourse. When a woman knows that any public statement she makes may be weaponized with a synthetic intimate image, the rational response is silence. This is a direct threat to democratic life.
Algorithmic hiring bias
AI tools used to screen job applications are in widespread use across Canadian industries. These tools learn from historical hiring data, and historical hiring data encodes historical discrimination. Amazon famously scrapped its AI hiring tool in 2018 after discovering it had systematically downgraded resumes that included the word "women's" and penalized graduates of all-women's institutions. The tool had learned, from the company's own past hiring patterns, to prefer men.
Amazon's case became public only because of the company's size and scale. The vast majority of AI hiring tools in use today have not undergone equivalent scrutiny. Canadian employers are not currently required to audit these tools for gender bias before deploying them.
Gender bias in medical AI
Most AI systems used in medical diagnosis were trained on datasets that significantly overrepresent male patients. This creates systems that are less accurate for women, particularly in areas where disease presents differently by sex. Heart disease, which presents differently in women than in men, is one well-documented example. Pain-related conditions, which are already undertreated in women relative to men, are another. A study published in The Lancet found that AI-based diagnostic tools showed significantly lower accuracy for women across multiple conditions.
When a medical AI is wrong, it doesn't make an individual error: it applies that error consistently to every patient who shares the underrepresented characteristic. The scale of deployment makes the scale of harm proportional.
AI-enabled harassment at scale
AI tools are being used to harass women online with an efficiency that has no historical precedent. Automated accounts generate hundreds of targeted messages per hour. AI voice synthesis creates threatening audio. Deepfakes are deployed as coercion: "comply, or we release this." For women who speak publicly on any platform, these are not hypothetical risks.
Canadian law has not kept pace with the scale or sophistication of AI-enabled harassment. Existing criminal harassment provisions were written for human perpetrators operating at human speed. The law does not clearly address automated abuse, and the platforms that host it are not currently held liable for facilitating it.
Labour disruption and the care-work gap
The occupations most exposed to AI automation are disproportionately held by women: administrative roles, data processing, customer service, healthcare support. At the same time, the unpaid care work that women disproportionately perform, including childcare, eldercare, and domestic labour, is not being automated. This creates a scenario where AI erodes paid employment for women while their unpaid obligations remain unchanged, deepening economic inequality rather than reducing it.
The core issue: Many of AI's most acute harms, including deepfake abuse, hiring bias, and medical misdiagnosis, fall disproportionately on women. These are not accidents of technology. They are the result of building systems on biased data, deploying them without equity audits, and failing to enforce existing legal protections at the speed the technology demands.
What to watch for in Canada
- Whether Canada's AI legislation requires mandatory gender-bias audits for AI hiring tools before deployment
- Whether Health Canada establishes equity requirements for AI diagnostic tools approved for clinical use
- Whether the Criminal Code is updated to address AI-generated non-consensual images at the scale they are now produced
- Whether Canada's Online Harms Act (Bill C‑63) creates meaningful platform accountability for AI-enabled abuse
Small business owners face AI disruption with fewer defences than large corporations
AI is reshaping competition in ways that disproportionately favour large, resource-rich organizations. Large enterprises can deploy AI across customer service, marketing, logistics, legal, and finance at a scale most small businesses cannot match. At the same time, small business owners face new AI-enabled threats they are often not equipped to defend against: sophisticated fraud, AI-generated competition, and dependency on platforms they do not control.
Competitive displacement
When a large company replaces hundreds of customer service representatives with an AI system, its cost structure changes fundamentally relative to competitors who cannot afford the same investment. When a major retailer uses AI to optimize supply chain, pricing, and inventory in real time, smaller competitors navigate with less information and slower decision cycles. AI is not inherently anti-competitive, but access to it is unevenly distributed, and the gap between what large enterprises and small businesses can deploy is widening.
For Canadian small business owners, this is already shaping competition in retail, financial services, hospitality, and professional services.
AI-generated competition
AI tools can now generate realistic product images, persuasive marketing copy, functional websites, and sophisticated software at near-zero marginal cost. For businesses in creative, communications, or professional services, AI-generated alternatives are already a competitive reality. A business that has charged a premium for quality copywriting, graphic design, or accounting support is now competing against AI that produces similar outputs for a fraction of the cost.
This does not mean these businesses are finished, because AI-generated work has meaningful limitations, and many clients still value human judgment, accountability, and relationships. But the market is shifting faster than most business owners anticipated.
AI-powered fraud targeting business
Business email compromise, fraudsters impersonating executives, suppliers, or banks to authorize wire transfers, has been significantly accelerated by AI. Criminals can now generate convincing impersonation emails, synthesize the voice of a CEO to authorize a transfer by phone, and personalize attacks at scale. Small businesses are disproportionately targeted because they typically have less robust security infrastructure and fewer internal controls than large enterprises.
AI has also made phishing attacks far harder to spot. Messages that previously required skilled writers are now generated instantly, personalized to the recipient, and free of the grammatical tells that once made phishing attempts identifiable.
Intellectual property and data privacy
Businesses that share proprietary information with AI tools, including customer data, strategic plans, internal documents, and trade secrets, may be inadvertently contributing that information to AI training datasets. Most major AI providers include language in their terms of service allowing them to use submitted content to improve their models, though terms vary and change without notice. Canadian privacy law (PIPEDA and provincial equivalents) places obligations on businesses around data sharing, but many business owners do not fully understand how their AI tool use interacts with these obligations.
Platform dependency and algorithmic risk
Most small businesses are heavily dependent on AI-powered platforms, including search algorithms, social media recommendation systems, and e-commerce rankings, over which they have no control, no transparency, and no contractual recourse when things change. A shift in Google's search algorithm or Amazon's ranking system can devastate a small business overnight, with no warning and no explanation. AI makes these platforms more powerful and their decisions less predictable.
The core issue: AI's economic benefits are disproportionately flowing to large, resource-rich organizations, while its risks, including fraud, platform dependency, competitive pressure, and legal uncertainty, fall on small businesses that are less equipped to navigate them. Canadian AI policy has not yet seriously addressed this asymmetry.
What to watch for in Canada
- Whether Canada's AI and competition policy addresses the concentration of AI capability among a handful of large platforms
- Whether BDC (Business Development Bank) and ISED programs help small businesses access and deploy AI competitively
- Whether Canada's privacy law reforms (Bill C‑27) create clear rules around business use of AI tools and customer data
- Whether the federal government supports Canadian-owned AI infrastructure to reduce dependency on US platforms
AI was built on artists’ work without consent, and now competes directly with them
The creative sector faces a structurally different challenge from what other industries are experiencing. AI systems were literally trained on artists' work, without permission, without compensation, and without the knowledge of the creators involved, and are now generating outputs that compete with the very artists whose work made them possible. This is not analogous to a technology that automates repetitive tasks. It is a situation where an industry's creative output was taken, industrialized, and turned against it.
Training data without consent or compensation
AI image generators, music AI, and writing tools were trained on enormous datasets scraped from the internet, including copyrighted artwork, published novels, recorded music, and photographs. In virtually none of these cases did the creators consent, receive payment, or even know their work was being used. Class-action lawsuits by artists, stock image companies, and news organizations are underway in the United States and the United Kingdom.
In Canada, copyright law has not been updated to address AI training data, and the federal government has not yet taken a clear position on whether training AI on copyrighted material without permission constitutes infringement. The Status of the Artist Act, which governs aspects of creator compensation in Canada, predates AI entirely.
Style replication and market disruption
AI tools can now convincingly replicate the distinctive visual style of living artists, sometimes by naming them explicitly in a prompt. This is possible because copyright law protects specific works but not artistic style. An artist who has spent years developing a recognizable aesthetic has no legal recourse against an AI that has absorbed that aesthetic and can reproduce it on demand, at zero cost, for anyone who asks.
Illustrators, graphic designers, and visual artists are seeing clients request "something in the style of [artist name]" using AI, work they would previously have commissioned from that artist directly. Stock image markets have already seen significant revenue declines as AI-generated images flood the market. The disruption is real and already happening.
Deepfakes of performers
AI voice synthesis and video generation are creating new categories of harm for musicians, actors, and public performers. Convincing vocal performances by specific artists can now be synthesized without their participation. Non-consensual "performances", such as an artist appearing to endorse a product they did not endorse or a deceased musician "releasing" new material, are technologically feasible today. The Canadian music industry raised these concerns with the federal government during Bill C‑11 (Online Streaming Act) consultations.
The legal framework for performers' rights in Canada, primarily the Copyright Act and performers' protection provisions, was not designed to address AI synthesis. Who owns a synthesized performance, and what consent is required to create one, has not been resolved in Canadian law.
The economics of creative work
Creative work has always been economically precarious for most practitioners. AI does not create that precarity, but it accelerates and deepens it. When AI can generate serviceable images for a marketing brochure, write adequate ad copy, or produce background music for a video in seconds, market prices for that tier of creative work come under downward pressure. Stock image markets, entry-level design work, and commercial writing are already seeing this compression.
AI and Canadian cultural sovereignty
Canada has long maintained cultural policies, including broadcasting requirements, content subsidies, and CanCon regulations, premised on the belief that a healthy democracy requires a domestic cultural voice. These policies were designed for a media landscape where the barrier to distribution was high. AI dramatically lowers production costs, making it trivially easy for American platforms to flood Canadian digital spaces with AI-generated content. Whether Canada's cultural policy frameworks are adequate to this challenge has not been seriously addressed.
Canada Council for the Arts, Telefilm Canada, the CBC, and provincial funding bodies will all face pressure to define their position on AI-generated content. Their decisions will shape whether a viable space for Canadian human creative work is maintained, or slowly ceded to platforms.
The core issue: Canadian artists did not consent to having their work used to train AI systems. They are now competing against systems built on their own creative output. Canadian copyright law, cultural policy, and performers' rights law have not been updated to address this reality. The federal government's response so far has been cautious and slow relative to the speed at which the market is moving.
What to watch for in Canada
- Whether the Copyright Act is amended to address AI training data, specifically whether training on copyrighted work without consent constitutes infringement
- Whether Canada Council for the Arts develops an explicit AI policy governing its funding decisions
- Whether Canada supports an international treaty on AI and creative rights (under discussion at WIPO)
- Whether Bill C‑11 regulations require platforms to fund Canadian content created by humans
- Whether Canada follows the EU in requiring disclosure when AI is used to generate published cultural content
Older Canadians are targeted by AI-powered fraud and left behind by AI-designed systems
Older Canadians face a distinctive combination of AI risks: they are the primary targets of AI-powered fraud schemes, they risk being excluded from services that are increasingly AI-dependent, and their health and care needs are being mediated by AI systems not designed with their situations in mind. At the same time, seniors are among the least represented in AI policy discussions and in the technical teams building these systems.
Voice cloning fraud and the grandparent scam
The "grandparent scam", a fraudster calling an older person pretending to be a grandchild in distress and urgently requesting money, has existed for years. AI voice cloning has made it vastly more convincing. A fraudster can now synthesize a convincing approximation of a specific person's voice from just a few seconds of audio available on social media. The call no longer sounds like a stranger: it sounds like family.
The Canadian Anti-Fraud Centre reported that Canadians lost over $554 million to fraud in 2023, with seniors disproportionately represented among victims of phone and digital fraud. As AI voice and video synthesis improves, the sophistication of these schemes will continue to increase. The technology is freely available; the legal frameworks for prosecuting its misuse often lag far behind.
Digital exclusion from AI-dependent services
AI is being embedded into systems Canadians depend on: healthcare portals, government benefits applications, banking, customer service, and communication platforms. As these systems become more AI-mediated, older Canadians who are less comfortable with digital interfaces, or who lack reliable internet access, risk being excluded from services they need and have a right to access.
There is a real risk of a two-tier service system: one for people who can navigate AI-mediated interfaces, and one, slower, less functional, and increasingly under-resourced, for those who cannot. Service Canada, the CRA, and provincial health authorities have all moved toward digital-first service models without consistently ensuring adequate non-digital alternatives for older Canadians.
Healthcare AI and older patients
AI diagnostic and decision-support tools are being deployed across Canadian healthcare. Many were trained on datasets that underrepresent older patients, who have more complex medical histories, more comorbidities, and different disease presentations than the younger, healthier populations that dominate most medical datasets. Growing evidence suggests that AI diagnostic tools perform less accurately for older patients than for younger ones.
Beyond accuracy, there is a care quality concern. AI is being used to automate decisions about care pathways, medication management, and service allocation in ways that reduce human judgment and the therapeutic value of human contact. For older patients, who often benefit most from consistent relationships with care providers, this shift carries real costs that do not appear in efficiency metrics.
Privacy and surveillance in eldercare
AI surveillance tools are being deployed in long-term care settings: motion sensors, cameras, and health monitors that continuously track the movement, sleep, and behaviour of residents. These tools are typically framed as safety measures. But they also represent continuous surveillance of some of the most vulnerable Canadians in settings where they have limited choice and limited capacity to give meaningful consent.
Who has access to this data, how long it is retained, whether it is shared with third parties, and what decisions it informs have not been adequately answered in most care settings. Canadian privacy law applies to these contexts, but enforcement is limited and few older Canadians in care have the resources to pursue complaints.
Algorithmic discrimination in financial services
AI systems used in credit scoring, insurance pricing, and fraud detection can discriminate against older adults in ways that are difficult to detect and contest. AI fraud detection may incorrectly flag ordinary financial behaviour by seniors, including large transfers to family members, irregular spending patterns, and unusual transaction timing, as suspicious, leading to account freezes or service disruptions. When these decisions are made by algorithms, there is often no transparent appeals process and no individual accountable for the outcome.
The core issue: Older Canadians are simultaneously among the most targeted by AI-enabled harm, including fraud, surveillance, and service exclusion, and among the least represented in the rooms where AI policy is being designed. An AI governance framework that does not specifically address the needs and vulnerabilities of older Canadians is a framework built for someone else.
What to watch for in Canada
- Whether federal agencies (CRTC, FCAC, OSFI) establish specific AI-related consumer protections for older Canadians
- Whether Canada's long-term care standards address AI surveillance and the consent rights of residents
- Whether Health Canada requires AI medical tools to demonstrate adequate performance across age groups before approval
- Whether Service Canada and provincial governments maintain accessible non-digital service pathways as they digitize
- Whether the Canadian Anti-Fraud Centre receives increased resources to address AI-powered fraud targeting seniors
AI in warfare is outpacing the rules designed to limit it
Autonomous weapons systems, AI-assisted targeting, and algorithmically-driven battlefield decision-making are no longer theoretical. They are being tested and deployed in active conflicts. For people who work in the military, defence policy, or national security, the implications go beyond operational effectiveness: AI is changing who makes lethal decisions, how fast those decisions happen, and whether anyone can be held accountable when things go wrong.
Rapid escalation and the shrinking window for de-escalation
Modern AI-enabled weapons systems can detect, classify, and respond to perceived threats faster than any human decision cycle allows. In a confrontation between two AI-capable adversaries, automated responses can occur before a commander is aware a triggering event has taken place. The time that has historically existed for communication, negotiation, and de-escalation, including a phone call between generals, a diplomatic back-channel, or a pause to verify intelligence, may not exist in a conflict where both sides are operating autonomous systems.
This is not a hypothetical risk. The 2020 Nagorno-Karabakh conflict saw AI-guided drone swarms used at battlefield scale for the first time, compressing engagement timelines dramatically. In a confrontation between major nuclear-armed powers, the margin between an automated response and an irreversible international incident could narrow to seconds. Systems designed to react faster than adversaries may, in aggregate, produce a situation that escalates faster than any human actor intended.
Civilian targeting errors at machine scale
AI targeting systems classify potential threats through pattern recognition, inferring combatant status from observed behaviour, location, communications signatures, vehicle type, and other data. This inference can be wrong. When a human soldier misidentifies a target, it is an individual error with individual consequences. When an AI targeting system is miscalibrated, it applies that error systematically to every person or object matching the flawed pattern.
In April 2024, Israeli investigative outlet +972 Magazine and Hebrew-language publication Local Call published a detailed investigation into the use of an AI system called "Lavender" by Israeli military intelligence. Based on interviews with current and former intelligence officers, the reporting found that Lavender generated a database of approximately 37,000 Palestinians flagged as suspected Hamas or Palestinian Islamic Jihad militants. Sources told the outlets that the system had a roughly 10% error rate, meaning around one in ten people marked as targets was likely a civilian with no militant affiliation. Across a list of 37,000 names, that error rate translates to an estimated 3,700 people wrongly designated for potential lethal targeting.
The human oversight described by soldiers was minimal. Sources said strike decisions were authorized in as little as 20 seconds per target, with soldiers describing their role as little more than a rubber stamp on the algorithm's output. The IDF reportedly approved up to 15 to 20 civilian deaths as acceptable collateral damage when targeting a low-ranking militant, and up to 100 or more for a senior commander. A companion system called "Where's Daddy?" was used to track targets to their family homes, meaning many were killed at night alongside their families, including children. The IDF has disputed some characterizations in the reporting, but has not denied the existence of either system.
International humanitarian law requires that each attack be individually assessed for proportionality and distinction between combatants and civilians. The Lavender reporting describes a process where that legal requirement was compressed to seconds and the core targeting decision was delegated to an algorithm with a known error rate. AI systems that generate target lists faster than meaningful human assessment can occur are in structural tension with the laws of armed conflict, regardless of how their operators are instructed to use them.
Expanding the definition of a legitimate target
International humanitarian law permits attacks on infrastructure that provides a direct military advantage: the doctrine of dual-use targeting. As military operations become more dependent on AI, the infrastructure that supports AI (data centres, cloud computing facilities, undersea communications cables, satellite uplinks, power grids) becomes arguable as a military target. Civilian populations who depend on the same infrastructure for hospitals, emergency services, water treatment, and communications bear the consequences of that targeting logic.
This is not a distant risk. Military planners in multiple countries are actively debating the legal and strategic status of commercial cloud infrastructure that also hosts defence workloads. If a data centre processes both civilian banking and military logistics, its status under IHL is genuinely ambiguous, and AI's integration into military systems is making that ambiguity more common, not less.
The accountability vacuum
Under international humanitarian law, military personnel are legally responsible for the lawfulness of the strikes they authorize. That accountability chain depends on a human making a decision. When an autonomous system selects a target and initiates engagement without meaningful human review, the chain of accountability breaks, and it does not simply transfer to the machine. It diffuses.
Is the engineer who wrote the targeting algorithm responsible? The procurement official who approved the system? The commander who deployed it to a theatre without fully understanding its parameters? The politician who authorized the defence program? In practice, full autonomy allows lethal force to be applied while accountability is spread so thin across so many actors that no individual can be held legally responsible for the outcome. For some governments, this diffusion of responsibility is not a problem to be solved, but an advantage to be preserved.
Canada's existing military law and the National Defence Act were not written to address autonomous weapons. The question of what constitutes lawful authorization of lethal force when an AI system is pulling the trigger has not been resolved in Canadian law.
Proliferation and lowered barriers to lethal force
Autonomous drone systems and AI-assisted weapons capabilities that required nation-state industrial capacity to develop a decade ago are now commercially available and rapidly decreasing in cost. Non-state armed groups, terrorist organizations, and criminal networks are beginning to deploy AI-assisted weapons systems. The proliferation of these capabilities to actors who have no obligation to comply with international humanitarian law, and no deterrable territory or population, changes Canada's security environment in ways that do not map cleanly onto existing defence doctrines.
Adversarial manipulation of AI systems
AI targeting and intelligence systems can be deliberately deceived. Adversarial inputs, carefully designed signals that cause AI systems to misclassify what they observe, are an active area of both research and operational planning. A targeting AI can potentially be manipulated into identifying civilian sites as military targets, or military targets as civilian sites. The reliability of AI systems under adversarial conditions, in environments specifically designed to defeat them, is a question that cannot be answered in testing environments and has serious implications for operational decisions made under that uncertainty.
The core issue: AI is being integrated into warfare faster than the legal, ethical, and strategic frameworks designed to govern it can adapt. The risk is not only that individual systems fail: it is that the cumulative logic of automated, high-speed, distributed lethal decision-making produces outcomes that no single actor intended and no existing legal framework can assign responsibility for. Canada has not yet legislated meaningful constraints on autonomous weapons or established what human authorization must exist before AI systems can initiate lethal force.
What to watch for in Canada
- Whether Canada supports a binding international treaty on Lethal Autonomous Weapons Systems (LAWS) at the UN Convention on Certain Conventional Weapons, as it did for landmines and cluster munitions
- Whether the National Defence Act is updated to define the minimum level of human authorization required before AI systems can initiate lethal force
- Whether Canada establishes mandatory accountability mechanisms for harm caused by AI systems deployed by the Canadian Armed Forces
- Whether Canada's defence procurement process requires independent ethical review of AI-enabled weapons systems before operational deployment
- Whether Canada develops a public policy position on the dual-use targeting of civilian AI infrastructure in armed conflict
Film workers are losing their likenesses, their labour, and their leverage to AI
The Canadian film and television industry has been at the centre of one of the most consequential labour confrontations with AI in any sector. The 2023 SAG-AFTRA strike in the United States, which lasted 118 days and was the longest in the union's history, was driven substantially by AI — specifically by studios' attempts to secure unlimited rights to scan and digitally reproduce performers' likenesses, replace background performers with AI-generated digital extras, and automate post-production work in ways that would reduce demand for editors, VFX artists, and crew. The strike set precedents that are still being interpreted. In Canada, similar negotiations are underway through ACTRA, IATSE, and other unions. The technology is moving faster than the collective agreements designed to govern it.
Actors and actresses: likenesses, synthetic performance, and the consent problem
The central AI issue for on-screen performers is not replacement, though that risk is real. It is ownership. AI voice synthesis can now replicate a specific actor's voice from a small sample of audio with results that are difficult to distinguish from a genuine recording. AI video generation can place a performer's likeness into scenes they never appeared in. And the contracts that govern most Canadian productions were written before either capability existed.
Studios have moved to include AI provisions in performer contracts that, in some cases, grant rights to scan and use an actor's digital likeness in perpetuity, across future projects, without additional compensation or approval. For established stars with leverage, these clauses are often negotiable. For mid-career and emerging performers who need the work, they frequently are not. The performer who accepts a one-time payment for a full digital scan may later discover their likeness being used in productions they never agreed to, cannot control, and earn nothing from.
AI casting tools are also being used to pre-screen auditions, analyzing facial expression, voice, and on-camera presence through algorithms whose criteria are opaque and whose biases have not been independently audited. A performer rejected by an AI screening tool has no recourse, no explanation, and often no knowledge that the decision was made by an algorithm rather than a human reader.
In Canada, the Copyright Act protects specific performances but does not clearly address whether an AI-synthesized reproduction of a performer's voice or likeness constitutes infringement. ACTRA's standard agreements are being updated in ongoing negotiations, but coverage is not uniform across productions, and non-union productions, which represent a significant share of Canadian screen output, are not subject to these protections at all.
Background performers: scanned once, replaced forever
Background performers, the thousands of Canadians who appear as crowd members, passersby, diners, office workers, and extras across Canadian film and television productions, face the most immediate displacement risk from AI. The technology already exists to populate large crowd scenes entirely with AI-generated figures, to replace background performers in post-production, and to scan a group of background performers once and reproduce their likenesses across unlimited future productions.
During the 2023 SAG-AFTRA strike, leaked studio proposals revealed that some production companies had sought the right to scan background performers' faces, bodies, and movements for a single day's pay, with the studio retaining unlimited ownership of the resulting digital model. That model could then be deployed in any future production, indefinitely, without additional consent or compensation. The deal would effectively convert a day's background work into a permanent, unpaid transfer of a person's entire physical identity.
For Canadian background performers, who are represented by ACTRA Background but work across a wide range of productions with varying levels of union coverage, the risk is acute. The Canadian film industry employs tens of thousands of background performers each year, primarily in Toronto, Vancouver, and Montreal. These are real economic livelihoods, and they are disproportionately held by new immigrants, racialized Canadians, and people who rely on background work as a primary income source. The displacement of background work is not an abstraction: it removes entry-level pathways into the industry and eliminates income for workers with limited alternatives.
The technical barrier to AI crowd generation is already low and falling. Productions that currently use 200 background performers for a crowd scene may, within a few production cycles, be able to generate that crowd entirely from AI, with no union agreement requiring consent, no payment, and no human presence on set.
Editors and VFX artists: automation from inside the craft
The disruption facing editors and visual effects artists is different in character from what performers face. It is less a question of replacement than of compression: AI tools are reducing the time required to complete tasks that have historically defined the craft, shrinking the size of the teams required, and changing the skill set that is valued and rewarded.
Rotoscoping, the painstaking frame-by-frame isolation of figures from backgrounds that was once a foundational VFX skill, can now be completed by AI tools in a fraction of the time it once required. AI upscaling and restoration tools have automated work that previously required specialist restoration artists. Scene detection, rough cut assembly, and dialogue cleaning, tasks that occupied significant portions of an editorial assistant's time, are being handled by AI with increasing reliability.
More disruptive than individual tool automation is the emergence of generative AI in the pre-production and production pipeline. AI is being used to generate concept art and pre-visualization sequences that studios previously commissioned from artists. AI de-aging and face replacement, used in productions including several major Marvel and Lucasfilm releases, reduces the VFX work required to achieve effects that once took large specialist teams months to complete. As studios increasingly turn to generative AI for environments, set extensions, and creature generation, the demand for the mid-level VFX generalists who have formed the backbone of Canadian VFX houses is under real pressure.
Canada has built a significant VFX industry, particularly in Vancouver and Toronto, partly through a combination of skilled labour and provincial tax credits that have attracted major productions from the US. That model depends on a supply of employable VFX workers. If AI systematically reduces the labour content of VFX work, the economic case for routing international productions to Canadian facilities weakens alongside it.
IATSE locals representing editors and VFX workers in Canada have begun incorporating AI provisions into collective agreements, including requirements for notice before AI tools are introduced to replace bargaining unit work. But enforcement is uneven, the technology evolves faster than bargaining cycles, and the rapid adoption of AI tools by major streaming platforms, which operate under different agreements than traditional broadcasters, creates gaps in coverage.
The core issue: Canadian film workers across all roles, from performers to post-production crew, are negotiating their relationship with AI technology without a clear legal framework to stand on. The Copyright Act does not address AI synthesis of performances. ACTRA and IATSE agreements are being updated but cover only a portion of the industry. Provincial and federal screen funding bodies have not yet established AI policies that condition public funding on performer consent or crew protections. The workers most at risk, background performers and mid-level VFX artists, have the least bargaining power and the least visibility in policy discussions.
What to watch for in Canada
- Whether ACTRA secures AI likeness protections equivalent to the SAG-AFTRA 2023 agreement — specifically prohibiting the scanning of background performers for a single day's pay with unlimited future use
- Whether the Copyright Act is amended to give performers a clear right to control and be compensated for AI reproductions of their voice or likeness
- Whether Telefilm Canada, the Canada Media Fund, and provincial screen agencies establish conditions on public funding that require productions to obtain explicit performer consent before using AI-generated likenesses
- Whether Canada's CanCon regulations are updated to clarify whether AI-generated content qualifies as Canadian content for funding and broadcasting purposes
- Whether IATSE agreements covering Canadian VFX houses include enforceable restrictions on AI replacing bargaining unit work without consultation and notice
- Whether the federal government's AI for All strategy includes sector-specific provisions addressing the film and television industry, which is one of Canada's largest cultural employers
What can you do?
Understanding how AI affects you is the first step. The Citizen Toolkit gives you concrete tools to make your voice heard: template letters to your MP and the Minister of AI, and step-by-step guidance on how to engage the policy process before decisions are made without you.