The use of technology to monitor workers is not new. Take, for example, the use of security cameras, keystroke loggers and screen-time monitoring. Artificial intelligence (AI), however, is introducing new forms of surveillance that are more intrusive and controlling.
Yet, despite AI’s growing adoption across workplaces, courts have issued surprisingly few decisions related to AI-enabled algorithmic management, especially as it relates to the psychosocial safety of workers.
Most Canadian occupational health and safety (OHS) legislation defines workplace hazards broadly, and courts have confirmed that psychosocial hazards fall within that scope. Canada’s national standard for psychological health and safety in the workplace, CSA Z1003-13, sets out an approach for identifying and controlling these hazards. Yet it offers no explicit guidance on where AI-enabled algorithmic management fits within that framework.
This is a notable gap as it is well documented that increased monitoring is a hazard, pressuring workers to work longer hours and contributing to psychological injuries such as stress or burnout. AI-enabled algorithmic management is similarly absent from provincial and territorial OHS regulations, leaving no explicit regulatory touchpoint at either the standards or legislative level.
Case law is a primary mechanism through which legal standards evolve. But without precedent, courts have had no opportunity to clarify whether existing OHS frameworks cover algorithmic monitoring, and regulators have had no judicial signal indicating where legislative gaps need to be closed. The gap is self-perpetuating: no legal precedent produces no legible harm, which produces no path for lawyers to litigate against worker harms, which limits the availability of legal precedent.
Legal frameworks weren’t built for this kind of technology. Privacy law is designed around concrete, identifiable data, not continuous learning systems whose outputs shift over time and cannot always be reconstructed. For a court assessing harm, the operative questions are no longer whether data is collected, but what the algorithm is doing, how that has changed since deployment and how harm could even be assessed.
Few legal cases have directly addressed AI’s effects on worker health and safety, though decisions addressing technological surveillance and monitoring more broadly do exist. In July, I led a comprehensive case law scan using WestLaw and Quicklaw, legal databases for academic use, which returned 602 legal cases and 1,834 tribunal decisions mentioning surveillance or monitoring alongside worker health and safety. Only a handful of these addressed technological monitoring and surveillance as a source of harm to workers; even among these, surveillance was consistently framed as a privacy issue rather than an occupational health and safety one.
Schindler Elevator Corp. illustrates this pattern: the company’s use of GPS and other electronic monitoring technology on company vehicles prompted a complaint alleging a violation of the Personal Information Protection Act. The British Columbia tribunal found no violation but required Schindler to revise its monitoring policy. The case turned entirely on consent and disclosure obligations, showcasing a legal framework responsive to identifiable data rather than an algorithm whose outputs evolve.
The lack of case law on algorithmic management and psychosocial hazards could be attributed to three explanations.
First, Canadian business and labour reporting has made algorithmic management appear widespread across the country. However, Statistics Canada data shows that in 2025, only 12 per cent of private businesses reported using AI. If AI-enabled algorithmic management represents only a fraction of that already-small figure, the underlying problem may be too small in scale to test whether the legal framework holds up.
Second, the impact of AI-enabled algorithmic management tools might be invisible to workers experiencing them. Tools can be embedded in day-to-day workplace software such as scheduling systems, productivity trackers and dispatch platforms. These uses may hide the underlying technological decision-making from the worker’s view. A worker who experiences an intensified pace of work may attribute this to a manager’s decision or a company policy without recognizing that an algorithm generated the outcome. This matters legally as much as empirically: a plaintiff must be able to name a specific cause of harm to bring a case at all. If workers cannot name AI as the source, they cannot frame a legal claim around it.
Third, AI-enabled algorithmic management may be more commonly deployed in workplaces characterized by precarity where workers lack the resources or power to litigate. Gig and platform work are the clearest cases. AI enabled systems assign jobs and rate performance, making monitoring a part of the job rather than incidental to it. These workers also are the least equipped to contest it. Most lack the internal grievance processes that might otherwise surface a complaint. Because platform workers often rely heavily on this income, challenging monitoring risks the job itself. And weak union presence leaves no collective mechanism to share the cost or risk of a test case. Uber v. Heller (2020), in which the Supreme Court of Canada only just cleared the way for drivers to challenge their classification as independent contractors, illustrates how recent and contested even baseline employment protections have been for this workforce. If workers in the most AI-intensive sectors are also the workers least equipped to access legal remedies, whether due to resource constraints or fear of retaliation, then the case law gap is not evidence that psychosocial harm from monitoring is rare. Rather, it may be evidence that the workers most exposed to that harm are the least able to bring it to court.
Left legally unaddressed, the psychosocial harms tied to surveillance and algorithmic management force workers to absorb stress, burnout and work intensification with no clear avenue for recourse. AI intensifies these harms in ways earlier forms of monitoring did not. A manager’s spot-check or a camera in a warehouse is a discrete, visible event; an algorithm that continuously works based on inferred patterns is not. Workers cannot see the system operating, let alone point to the moment it caused them harm, making it hard to name and harder to litigate.
The absence of Canadian case law is not proof the harm is minor. It reflects a legal framework built for a different kind of technology; a workforce that cannot see the systems monitoring it; and workers who are unable to challenge those systems when they are the most exposed.
Closing this gap will require deliberate intervention. Regulators should not wait for a test case to clarify that psychosocial hazard provisions extend to AI-enabled algorithmic management; legislators can build explicit AI-specific obligations into OHS frameworks now, rather than after harm has already accumulated.
Without that intervention, workers will keep absorbing the psychosocial cost of a technology the law has not yet named.
This article is written with assistance and guidance from Dr. Arif Jetha, Scientist at Institute for Work and Health, and Dr. Victoria Arrandale, Assistant Professor at the University of Toronto’s Dalla Lana School of Public Health.
