ABSTRACT
Background
Artificial intelligence (AI) can modernize occupational health and safety (OHS) practice and provide solutions to the most complex health and safety challenges. Empirical data on firm‐level AI utilization in OHS practice remain limited. The objective of this study was to examine AI use for OHS and firm‐level descriptive and OHS characteristics associated with AI use.
Methods
A total of 810 OHS professionals in British Columbia and Ontario, Canada were surveyed in the summer of 2024. Surveys asked about firm‐level AI use for OHS and items asked about descriptive and OHS characteristics. Participants were also asked about perceived AI concerns and OHS impact. A multivariate logistic regression model was fitted to examine factors associated with firm‐level AI use for OHS.
Results
In total, 29% reported firm‐level AI use for OHS. Larger‐sized firms and those with hybrid work arrangements had a greater odds of AI use for OHS. Also, firms with high workplace hazard exposure had a greater odds of AI OHS use. More positive perceptions of AI's impact on OHS were associated with firm‐level AI use for OHS.
Conclusions
AI use for OHS may be concentrated among hazardous firms and those with the conditions to support technological adoption. Research examining AI's effectiveness in OHS settings is needed to guide evidence‐based implementation in occupational health practice.
Keywords: artificial intelligence, occupational health, surveys and questionnaires, workers, workplace
1. Introduction
Artificial intelligence (AI) holds promise to enhance occupational health and safety (OHS) practice. While there is broad enthusiasm regarding the role AI can play in transforming the labor market, there remains limited information on the extent to which AI is used by firms to advance OHS. We present findings from a survey of OHS professionals in two Canadian provinces to examine firm‐level AI use for OHS and examine the characteristics that differentiate AI‐using firms from non‐AI using firms.
AI broadly refers to the use of computing machines to solve problems traditionally requiring human intelligence and may be used to detect patterns, make predictions and decisions, automate physical and cognitive job tasks, or generate outputs (e.g., images, text, video) [1, 2, 3, 4]. AI applications exist in almost all industries including, but not exclusive to, healthcare (e.g., diagnostic tools), manufacturing (e.g., intelligent cobots), finance (e.g., automated stock trading platforms), and transportation (e.g., self‐driving vehicles). A recent global survey of organizational representatives across a variety of industries found that almost three‐quarters indicated that their firms had adopted AI in some capacity and over two‐thirds indicated that they would increase their investments in AI over the next 3 years [5]. Also, Canadian data show that larger firms and those based in regions where there is a strong infrastructure to foster technological adoption represent characteristics that are associated with AI utilization [6]. The growing affordability of computational power and the increased availability of large amounts of high‐dimensional data have made AI more accessible and applicable across various industries, suggesting continued and widespread adoption of the technology [7, 8].
Leveraging new technology has been a strategy taken by OHS professionals to enhance OHS risk identification and mitigation strategies [9]. The ability to quickly analyze, find patterns, and make predictions from large amounts of data sets AI apart from past OHS technological applications. Recent literature on AI in the field of OHS highlights the technology's potential to identify hazards, eliminate unsafe conditions, promote safety behaviors, and produce in‐depth insights that support organizational decision‐making [10, 11, 12]. Accordingly, the choice to use AI for OHS may reflect a workplace's motivation to creat innovative strategies to reduce workplace injury and illness. To date, we currently know very little about the extent to which firms are using AI to protect worker safety and promote improved health outcomes. A recent systematic review shows very limited evidence for the efficacy of AI tools for health and safety outcomes [13]. Despite AI's promise, there is a growing discourse on AI risks, including privacy concerns associated with large data requirements and AI mistrust [14. Potential risks that may emerge with the adoption of the technology have important implications for OHS [13].
Our study aimed to describe the use of AI for OHS within firms in British Columbia and Ontario, Canada. Through a survey of OHS professionals, we addressed three primary research objectives: (1) to estimate the prevalence of firms using AI for OHS; (2) to examine how OHS professionals perceive AI for OHS and how their perceptions are associated with firm‐level AI use for OHS; and (3) to determine firm‐level and OHS characteristics associated with AI use for OHS.
2. Methods
2.1. Participants and Recruitment Approach
Participants were recruited from British Columbia and Ontario which are amongst the largest Canadian labor markets, a large number of workers and a similar composition of workers employed in different industries. To be eligible, participants had to self‐identify that they were employed in a role responsible for OHS, ≥ 18 years of age, and fluent in English. Participants were recruited from April to June 2024 through a survey research firm that maintains a panel of 400,000 members whose sociodemographic characteristics reflect the broader Canadian population [15]. Panel quality is maintained by the research firm through several strategies including a double opt‐in process and the removal of panel participants with invalid emails or erroneous IP addresses. To ensure data quality, members of the research team monitored survey completion time and examined floor and ceiling effects in participant responses.
Panel members who met eligibility criteria were sent an email invitation and study information by the research firm which asked them to confirm that they fit the inclusion criteria. Interested participants were then sent study consent information to review and electronically sign. Once informed consent was obtained, participants could access the survey. Utilizing the panel, we were able to identify 3199 eligible potential participants who received the study invitation. From those, 810 completed the survey. Surveys were administered using an online platform and took 15 min to complete. Participants were remunerated for their time through the survey research firm using points that could be redeemed for gift cards or cash. To protect participant confidentiality, no personal identifying information was collected, and participants were not asked to identify the specific firm in which they worked. All data were presented at an aggregate level and stored on an encrypted digital folder. The research protocol was reviewed by the lead investigator's research ethics board.
2.2. Survey Measures
The survey took a broad approach to examine any AI use for OHS and perceptions regarding AI. Prior to answering questions on AI, participants were provided with the following prompt to anchor their responses: “In this next section, we will ask you about your thoughts and perceptions regarding the use of AI as a tool to improve the health and safety of workers. When we refer to AI, we are referring to any technology that uses computing machines to solve problems traditionally requiring human intelligence, including detecting patterns, making predictions and decisions, or optimizing processes. Often, AI is used in conjunction with other types of technology like robotics, mobile devices, sensors, or computer software. Some examples of AI include chatbots, fraud detection software, software/cameras to identify flaws, prediction software, and technology that can anticipate when machines need to be serviced.”
2.2.1. Outcome Variable: Firm‐Level AI Use for OHS
An item developed for this survey asked the question, “To the best of your knowledge, does your company use AI for worker health and safety purposes?” Participant response was collected using a 5‐point Likert scale (1 = not at all; 5 = a great deal). Responses were categorized into AI‐using firms for OHS (any AI use) and non‐AI‐using firms for OHS. Participants indicating that they “didn't know about their firm's AI use for OHS” was a third category.
2.2.2. Firm‐Level Characteristics
Survey items asked about firm size (number of employees), work location (primarily at the workplace, primarily remote work, or a hybrid work arrangement), gender breakdown of workers (mostly women workers, mostly men workers, or a mix), and union representation of at least some workers (yes/no). Based on industry classification, firms were categorized as primary (e.g., mining, agriculture), goods‐producing (e.g., manufacturing), or sales and service industries (e.g., finance, healthcare).
2.2.3. Firm‐Level OHS Characteristics
Two previously validated measures were adapted to describe firm‐level OHS characteristics [16, 17, 18].
2.2.3.1. Adequacy of OHS Policies and Practices
Using seven items, participants were asked about the adequacy of different OHS policies and practices within their firms (e.g., OHS training, OHS communication, accident investigation protocols). For each item, participants rated their level of agreement on a 5‐point Likert scale (1 = strongly disagree; 5 = strongly agree). OHS policies or practices where there was agreement or strong agreement were categorized as adequate. The number of adequate OHS policies and practices was summed.
2.2.3.2. Workplace Hazards at a High Frequency
Using ten items, participants rated the frequency to which workers in their firm might be exposed to physical (e.g., repetitive tasks, heavy lifting), psychosocial (e.g., bullying or harassment), or chemical hazards, or unsafe working situations (e.g., working from heights) on a 7‐point Likert scale (0 = never; 6 = every day). Each hazard was characterized as high (exposure every 3 months/every month/every week/every day) or low hazard (exposure never/once a year/every 6 months). The number of high‐hazard exposures was summed and categorized as low (0–1 high hazards), moderate (2–4 high hazards), or high ( 5 high hazards).
2.2.4. AI Perceptions
Two measures created for this survey asked participants about their perceptions of AI in OHS.
2.2.4.1. AI Concerns
Using nine items, participants rated the extent to which they had concerns about AI including its cost, a lack of understanding of AI and its challenges, high technical expertise requirements, data needs, ethical concerns, privacy concerns, preference for traditional OHS strategies, and mistrust (see Appendix B, Table SI). Participant response was collected on a 5‐point Likert scale (1 = not at all; 5 = a great deal). Cronbach's α for AI concern items was 0.92 indicating high reliability. A mean total AI concern score was developed.
2.2.4.2. AI Impact
Four items asked about the perceived impact of AI on OHS outcomes including its ability to create a safer workplace, address OHS costs, address psychological injuries at work, and address physical injuries at work (see Appendix B, Table SII). Participant response was collected using a 3‐point Likert scale (1 = not at all; 3 = a great deal). Cronbach's α for AI impact items was 0.87 indicating good reliability. A mean AI impact score was developed.
2.3. Analysis
Descriptive statistics (frequencies, percentages, means, and standard deviations) were used to build a profile of firm‐level and OHS characteristics and participant perceptions regarding AI. Differences in firm‐level characteristics and perceived AI concerns were compared according to whether a participant reported that their firm used AI for OHS, did not use AI for OSH, or did not know if their firm used AI using bivariate statistics including analysis of variance (ANOVA) and χ 2 tests. Univariate and multivariate logistic regression analyses estimated odds ratios (ORs) and 95% confidence intervals (95% CIs) of the relationship between firm‐level descriptive characteristics and OHS characteristics and participant perceptions regarding AI and whether a firm used AI for OHS. Factors significantly associated with AI usage in the univariate model were carried over to the final multivariate model. Participants who did not know whether their firm used AI for OHS were excluded from univariate or multivariate models. Hosmer and Lemeshow Goodness‐of‐Fit tests were conducted to determine model fit (p < 0.05). Analyses were performed using SAS software (Version 9.4) [19]. The STROBE guidelines were used to inform the development of this manuscript (see supplement STROBE) [20].
3. Results
A total of 810 OHS professionals completed the survey, representing supervisors, senior managers, or executives with OHS responsibilities (61%), OHS practitioners (12%), human resource professionals with OHS responsibilities (10%), union representatives (3%), or other roles with OHS responsibilities (14%). Over a quarter of participants reported firm‐level AI use for OHS (29%), 54% indicated that their firms did not use AI for OHS, and 17% did not know if their firms used AI for OHS. Participants indicated several reasons for using AI for OHS including to identify and eliminate hazards (27%), monitor safety behaviors (22%), promote health‐enhancing behaviors (19%), prevent psychological injuries (15%), prevent physical injuries (10%), and for disability management (7%). Firm‐level descriptive and OHS characteristics and participant AI perceptions are summarized in Table 1 according to AI use for OHS.
Table 1.
Firm‐level descriptive characteristics, OHS characteristics, and participant AI perceptions for the full sample and compared according to AI use for OHS.
| Full sample (N = 810) | AI‐using firms for OHS (N = 231) | Non‐AI‐using firms for OHS (N = 439) | Don't know if AI is used for OHS (N = 140) | ||
|---|---|---|---|---|---|
| Variable | N (%)/Mean ± SD | N (%)/Mean ± SD | N (%)/Mean ± SD | N (%)/Mean ± SD | p* |
| Firm‐level descriptive characteristics | |||||
| Industry | 0.7147 | ||||
| Primary industries | 87 (10.7) | 27 (11.7) | 48 (10.9) | 12 (8.6) | |
| Goods‐producing industries | 568 (70.1) | 161 (69.7) | 302 (68.8) | 105 (75.0) | |
| Sales and service industries | 152 (18.8) | 42 (18.2) | 87 (19.8) | 23 (16.4) | |
| Province | 0.0107 | ||||
| British Columbia | 315 (38.9) | 71 (30.7) | 184 (41.9) | 60 (42.9) | |
| Ontario | 495 (61.1) | 160 (69.3) | 255 (58.1) | 80 (57.1) | |
| Firm size (number of employees) | 0.0001 | ||||
| Small ( 99) | 377 (46.5) | 82 (35.5) | 262 (59.7) | 33 (23.6) | |
| Medium (100–500) | 169 (20.9) | 54 (23.4) | 89 (20.3) | 26 (18.6) | |
| Large ( 500) | 249 (30.7) | 94 (40.7) | 82 (18.7) | 73 (52.1) | |
| Union representation | 0.0001 | ||||
| Yes | 293 (36.2) | 97 (42.0) | 119 (27.1) | 77 (55.0) | |
| No | 498 (61.5) | 125 (54.1) | 314 (71.5) | 59 (42.1) | |
| Gender breakdown at workplace | 0.0001a | ||||
| Mostly men | 127 (15.7) | 34 (14.7) | 84 (19.1) | 9 (6.4) | |
| Mostly women | 164 (20.3) | 33 (14.3) | 104 (23.7) | 27 (19.3) | |
| Mix of men, women, and others | 236 (29.1) | 85 (36.8) | 120 (27.3) | 31 (22.1) | |
| Don't know | 272 (33.6) | 77 (33.3) | 126 (28.7) | 69 (49.3) | |
| Work location | 0.0005 | ||||
| Primarily at workplace | 468 (57.8) | 117 (50.7) | 281 (64.0) | 70 (50.0) | |
| Primarily working remotely | 95 (11.7) | 26 (11.3) | 53 (12.1) | 16 (11.4) | |
| Hybrid work arrangement | 247 (30.5) | 88 (38.1) | 105 (23.9) | 54 (38.6) | |
| Firm‐level OHS characteristics | |||||
| Adequacy of OHS policies and practices | 0.0300 | ||||
| 0 | 76 (9.4) | 18 (7.8) | 44 (10.0) | 14 (10.0) | |
| 1–4 | 172 (21.2) | 47 (20.4) | 98 (22.3) | 27 (19.3) | |
| 5–6 | 197 (24.3) | 43 (18.6) | 123 (28.0) | 31 (22.1) | |
| 7 | 65 (45.1) | 123 (53.3) | 174 (39.6) | 68 (48.6) | |
| Workplace hazards at a high frequency | 0.1760 | ||||
| Low (0–1) | 212 (26.2) | 53 (22.9) | 123 (28.0) | 36 (25.7) | |
| Moderate (2–4) | 313 (38.6) | 83 (35.9) | 170 (38.7) | 60 (42.9) | |
| High ( 5) | 280 (34.6) | 94 (40.7) | 144 (32.8) | 42 (30.0) | |
| Participant perceptions of AI | |||||
| AI concerns | 3.26 ± 1.03 | 3.33 ± 0.86 | 3.14 ± 1.13 | 3.57 ± 0.83 | 0.0004 |
| AI impact | 1.83 ± 0.59 | 2.11 ± 0.58 | 1.69 ± 0.57 | 1.82 ± 0.51 | 0.0001 |
* χ 2 analyses (categorical variables) or ANOVA (continuous variables) were used to examine differences.
a Participant gender differences measured using Fisher's exact test due to small cell size for certain categories.
Most participants were from firms in goods‐producing industries (70%) and fewer participants were from firms in sales and service (19%) or primary industries (11%). Close to two‐thirds of participants worked in firms based in Ontario (61%). Just under half of participants indicated that their firms were small‐sized ( 99 employees; 47%), 20% indicated that their firms were medium‐sized (100–500 employees), and 31% indicated their firms were large‐sized ( 500 employees). Over one‐third of participants (36%) indicated at least some union representation and 29% indicated that their firms included a mix of women and men workers. Over half indicated that their firms required workers to be primarily at the workplace (58%) and 30% indicated a hybrid work arrangement. Descriptively, a greater proportion of AI‐using firms were based in Ontario and were of large‐ or medium‐sized when compared to firms not using AI for OHS or those where the participant did not know if their firm used AI for OHS. Also, a greater proportion of AI‐using firms were unionized, included a mix of women and men workers, and offered hybrid work arrangements.
A greater proportion of AI‐using firms were characterized by the adequacy of all seven OHS policies and practices (53%) when compared to non‐AI‐using firms (40%). In contrast, a greater proportion of non‐AI‐using firms (10%) indicated that no OHS policies or practices were adequate when compared to AI‐using firms (8%). A breakdown of the various OHS policies and practices and the frequency to which they were reported as being adequate are provided in Appendix A, Table SI.
Frequency of workplace hazard exposure also differed according to firm‐level AI use for OHS. High‐hazard firms were more likely to be AI‐using firms for OHS (41%) than non‐AI‐using firms for OHS (33%). Conversely, low‐hazard firms were more likely to be non‐AI‐using firms for OHS (28%) than AI‐using firms for OHS (23%). In Appendix A, Table SII, we describe the frequency of specific workplace hazards according to whether a firm used AI for OHS.
Participants from firms using AI for OHS reported higher perceived concerns regarding AI (mean score = 3.33, SD = 0.86) when compared to those from non‐AI‐using firms for OHS (mean score = 3.14, SD = 1.13). Described in Table 2 are the most highly rated AI concerns, which included a loss of privacy, high data requirements, mistrust toward AI, ethical concerns regarding automated decision‐making, and a preference for traditional OHS approaches.
Table 2.
Perceived concerns regarding the use of AI for OHS among all study participants and compared according to AI use for OHS.
| Full sample (N = 810) | AI‐using firms for OHS (N = 231) | Non‐AI‐using firms for OHS (N = 439) | Don't know if AI is used for OHS (N = 140) | ||
|---|---|---|---|---|---|
| Mean ±SD | Mean ±SD | Mean ±SD | Mean ±SD | p * | |
| Loss of privacy related to data collection | 3.47 ± 1.32 | 3.56 ± 1.02 | 3.32 ± 1.48 | 3.88 ± 1.18 | 0.0005 |
| Data needed for AI | 3.32 ± 1.33 | 3.26 ± 1.15 | 3.25 ± 1.45 | 3.76 ± 1.13 | 0.0050 |
| Mistrust toward AI | 3.32 ± 1.30 | 3.26 ± 1.14 | 3.26 ± 1.42 | 3.65 ± 1.09 | 0.0370 |
| Ethical concerns related to AI and automated decision‐making | 3.29 ± 1.34 | 3.27 ± 1.18 | 3.17 ± 1.48 | 3.80 ± 0.97 | 0.0005 |
| Preferences for traditional OHS approaches | 3.29 ± 1.26 | 3.42 ± 1.05 | 3.17 ± 1.40 | 3.54 ± 1.02 | 0.0147 |
| Technical expertise needed to adopt AI | 3.27 ± 1.30 | 3.37 ± 1.16 | 3.17 ± 1.41 | 3.49 ± 1.11 | 0.0537 |
| Cost of implementing AI tools | 3.08 ± 1.36 | 3.54 ± 1.10 | 2.75 ± 1.46 | 3.27 ± 1.21 | < 0.0001 |
| Insufficient understanding of AI | 2.98 ± 1.29 | 3.21 ± 1.02 | 2.77 ± 1.43 | 3.24 ± 1.13 | 0.0002 |
| Insufficient understanding of AI challenges | 2.99 ± 1.31 | 3.27 ± 1.15 | 2.81 ± 1.42 | 3.03 ± 1.08 | 0.0004 |
Differences between AI use determined using analysis of variance (ANOVA).
Participants from firms using AI for OHS held stronger perceptions that AI would have a positive impact on OHS outcomes (mean = 2.11, SD = 0.58) when compared to participants from non‐AI‐using firms (mean = 1.69, SD = 0.57). Most highly rated AI impacts included its ability to create a safer workplace and reduce OHS costs (see Table 3).
Table 3.
Perceived impact of AI among all study participants and compared according to AI use for OHS.
| Full sample (N = 810) | AI OHS‐using firms (N = 231) | Non‐AI‐using firms for OHS (N = 439) | Don't know if AI is used for OHS (N = 140) | ||
|---|---|---|---|---|---|
| Mean ±SD | Mean ±SD | Mean ±SD | Mean ±SD | p * | |
| Creating safer workplaces | 1.91 ± 0.66 | 2.22 ± 0.58 | 1.75 ± 0.66 | 1.96 ± 0.59 | < 0.0001 |
| Addressing the costs associated with OHS | 1.91 ± 0.70 | 2.11 ± 0.67 | 1.80 ± 0.71 | 1.97 ± 0.65 | < 0.0001 |
| Total perceived AI impact on OHS | 1.83 ± 0.59 | 2.11 ± 0.58 | 1.69 ± 0.57 | 1.82 ± 0.51 | < 0.0001 |
| Addressing psychological injuries | 1.75 ± 0.73 | 2.08 ± 0.74 | 1.62 ± 0.71 | 1.67 ± 0.61 | < 0.0001 |
| Addressing physical injuries | 1.73 ± 0.71 | 2.02 ± 0.70 | 1.60 ± 0.68 | 1.69 ± 0.67 | < 0.0001 |
Differences between AI use determined using analysis of variance (ANOVA).
A final multivariate logistic regression model was fitted to estimate the odds of firm‐level AI use for OHS (Table 4) and was informed by univariate analyses which examined individual effects (Appendix A, Table SIII).
Table 4.
Multivariate logistic regression model examining firm‐level likelihood of AI use for OHS.
| OR (95% CI) | |
|---|---|
| Firm‐level descriptive characteristics | |
| Industry | |
| Primary Industries | 1.01 (0.49, 2.08) |
| Goods‐producing industries | 0.96 (0.57, 1.62) |
| Sales and service industries | 1.00 |
| Province | |
| Ontario | 1.68 (1.11, 2.56) |
| British Columbia | 1.00 |
| Firm size (number of employees) | |
| Medium (100–500) | 1.69 (1.01, 2.81) |
| Large ( 500) | 2.45 (1.44, 4.17) |
| Small ( 99) | 1.00 |
| Union representation | |
| Yes | 1.40 (0.89, 2.21) |
| No | 1.00 |
| Gender breakdown at work | |
| Mostly men | 1.25 (0.63, 2.49) |
| Mix of men, women, and others | 1.59 (0.86, 2.95) |
| Don't know | 1.13 (0.60, 2.13) |
| Mostly women | 1.00 |
| Work location | |
| Primarily working remotely | 1.17 (0.61, 2.25) |
| Hybrid work arrangement | 1.79 (1.14, 2.80) |
| Primarily at workplace | 1.00 |
| Firm‐level OHS characteristics | |
| Adequacy OHS policies and practices | |
| 1–4 | 1.16 (0.52, 2.59) |
| 5–6 | 1.00 (0.45, 2.23) |
| 7 | 1.84 (0.86, 3.92) |
| 0 | 1.00 |
| Workplace hazards at a high frequency | |
| Moderate (2–4) | 2.02 (1.19, 3.45) |
| High (5 or more) | 3.28 (1.87, 5.74) |
| Low (0–1) | 1.00 |
| Participant perceptions of AI | |
| AI concerns | 0.82 (0.66, 1.02) |
| AI impact | 3.33 (2.27, 4.86) |
Note: Analysis excludes participants who indicated not knowing about firm AI use for OHS (n = 584).
Abbreviations: OR = odds ratios; CI = confidence intervals.
According to the model, firms based in Ontario were significantly more likely to report AI use for OHS when compared to firms based in British Columbia (OR = 1.68, 95% CI 1.11, 2.56). Larger‐ (OR = 2.45, 95% CI 1.01, 2.81) and medium‐sized firms (OR = 1.69, 95% CI 1.01, 2.81) had a greater odds of AI use for OHS compared to small‐sized firms. Firms with hybrid work arrangements had 81% greater odds of AI use for OHS compared to firms with the requirement for workers to be at the primary worksite (OR = 1.81, 95% CI 1.15, 2.83). Firms with moderate (OR = 2.02, 95% CI 1.19, 3.45) and high exposure to workplace hazards at a high frequency (OR = 3.28, 95% CI 1.87, 5.74) were more likely to report AI use for OHS when compared to firms with low workplace hazard exposure. Participants who held greater perceptions of the impact of AI on OHS outcomes were more likely to be based at a firm that used AI for OHS (OR = 3.33, 95% CI 2.27, 4.86).
4. Discussion
AI is increasingly prominent within Canadian workplaces. Our study provides a novel snapshot of firm‐level AI use for OHS. We observed that just under a third of participants indicated AI use for OHS. Firm size and the hazardousness of workplaces represented characteristics associated with greater odds of AI use for OHS. Participants who held greater perceptions of the utility of AI for OHS were more likely to indicate that their firms used AI for OHS. Findings highlight the need for ongoing research that seeks to examine how the technology is being used for health and safety purposes and its effectiveness in addressing workplace injuries or illnesses.
Just under a third of participants indicated firm‐level AI use for OHS. In comparison, past studies show that close to three‐quarters of firms may be adopting AI to enhance workplace productivity [21]. AI may be an underutilized technology for OHS. Interestingly, we found that 17% of participants did not know whether their firms used AI for OHS, which could reflect a lack of awareness among OHS professionals over whether they are encountering AI within their professional roles. At the same time, our results suggest AI adoption may create disparities in safety between organizations. Firms not using the technology may limit their ability to leverage the technology to protect workers from OHS risks. Given the powerful potential of the technology, there is an ongoing need to identify ways in which OHS practitioners may use AI to support a range of OHS activities, help address workplace injuries, and reduce workplace health and safety disparities. Our survey asked about AI applications broadly to capture the range of ways AI might be used for OHS practice. Future research should probe how specific AI applications are used to promote health and safety such as those that use machine learning or generative AI. The need for this study is underscored by technological advancements which are expected to accelerate the development and availability of AI‐powered OHS tools [5].
Organizational characteristics may shape the conditions to which AI can be deployed for OHS purposes. Aligning with previous research examining factors determining AI uptake for business purposes, we show that larger firms and those that offer hybrid work arrangements may be more likely to utilize AI for OHS [6]. It may be that larger firms have more resources to allocate toward OHS innovation and are more likely to invest in new digital tools to tackle complex health and safety challenges. Also, firms able to offer hybrid work arrangements may have an existing infrastructure (e.g., availability of specialized software, policies, and practices) that may make AI adoption more feasible and may also be more accustomed to adopting new technologies to foster virtual collaboration [22]. It is important to highlight that in Canadian contexts, such as Ontario and British Columbia where the study was conducted, a large proportion of employers are small‐ and medium‐sized or require workers come into the physical worksite [23, 24]. Accordingly, a large number of workers may be less likely to be exposed to AI and its impact on OHS. Additional research is warranted to examine how the technology may be applied to health and safety practices among firms that range in size, work location, and other characteristics.
Importantly, firms that were characterized as being high‐hazard were more likely to use AI for OHS. High‐hazard organizations could be among the most safety‐sensitive and may be more likely to look toward new technology to protect the health, safety, and well‐being of workers [25]. The ability for AI‐powered systems to automate physical and cognitive roles within the workplace may mean that the technology may be able to mitigate or eliminate a range of hazardous working situations [4]. At the same time, a recent systematic review shows a paucity of research on the measurable impact of AI on worker injury and illness outcomes [13]. To build on our findings and address knowledge and practice gaps, there is a need to determine how AI may be used to address a broad range of physical, cognitive, and psychosocial hazards and study its impact on OHS outcomes. Additionally, qualitative research may help to better understand how AI systems are being used in practice to promote health and safety and better understand the experiences of OHS professionals as they engage with the technology.
Unsurprisingly, participants from firms that used AI for OHS were also more likely to hold positive perceptions of the potential benefits of AI for health and safety. Our study could suggest that participants from AI‐using firms have a greater awareness of both the advantages and disadvantages of the technology. Participants from AI‐using firms may also have greater preparedness to navigate the different considerations that may emerge when adopting an AI system. We underscore the need for all OHS leaders, especially those in firms not currently using AI, to build AI literacy so that they are more fully able to understand the strengths and limitations of the technology and contribute to organizational decisions on the design and adoption of AI at work. Greater AI literacy will also enable OHS professionals to identify potential harms associated with the technology to better protect workers.
There are two study limitations to note. Currently, without direct observations, it is difficult to measure AI utilization. We relied on self‐reports of AI use within a participant's firm and we were unable to validate whether AI is used and how the technology may shape OHS outcomes. Additional research within workplaces may enhance an understanding of AI use and its impact on workplace injury and illness outcomes. Despite efforts to recruit a diverse sample, our study sample may have reflected limited variation in terms of industry which may affect the generalizability of our findings. Despite these limitations, our study is one of the first to describe firm‐level AI utilization for health and safety and draw on insights from a relatively large number of OHS professionals.
5. Conclusion
AI is reshaping the working world and may present benefits and challenges for OHS. Through findings of a survey of OHS professionals we describe AI use for health and safety purposes. Our findings may suggest that AI is underutilized for OHS purposes. We highlight the context in which AI may be most used, specifically among large and more hazardous firms, as well as those offering hybrid work arrangements. Amidst the growth of AI tools, ongoing research is needed to examine the role of AI technology in tackling a range of OHS challenges and further understand how it could pose benefits and harms to different firms.
Author Contributions
Conception or design of the work: Arif Jetha, Aviroop Biswas, Cameron Mustard, Maxwell J. Smith, and Peter M. Smith. Acquisition, analysis, or interpretation of data for the work: Arif Jetha and Hyunmi Lee. Drafting the work or revising it critically for important intellectual content: all authors. Final approval of the version to be published: all authors. Agreement to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved: all authors.
Ethics Statement
The work was performed at the Institute for Work & Health. The research protocol was approved by the lead investigator's Ethics Board, the University of Toronto's (U of T) Human Research Ethics Unit (HREU), REB #46114.
Consent
Written participant informed consent was obtained from participants prior to their participation in the research.
Conflicts of Interest
Several members of the authorship team (A.J., A.B., C.M., P.M.S.) are employees of the Institute for Work & Health (IWH), which is supported through funding from the Ontario Ministry of Labour, Immigration, Training and Skills Development (MLITSD). The analyses, conclusions, opinions, and statements expressed herein are solely those of the authors and do not reflect those of the MLITSD; no endorsement is intended or should be inferred. The rest of the authors have no competing interests.
Supporting information
Supplementary Information
Acknowledgments
Grant sponsors: Ontario Workplace Safety and Insurance Board; Grant Number: JETH2022; WorkSafeBC.
Institution at which the work was performed: Institute for Work & Health, Toronto, Canada.
References
- 1. Russell S. J. and Norvig P., Artificial Intelligence: A Modern Approach (Pearson, 2016). [Google Scholar]
- 2. Acemoglu D. and Restrepo P., “Tasks, Automation, and the Rise in US Wage Inequality,” Econometrica 90, no. 5 (2022): 1973–2016. [Google Scholar]
- 3. Brynjolfsson E. and Mitchell T., “What Can Machine Learning Do? Workforce Implications,” Science 358, no. 6370 (2017): 1530–1534. [DOI] [PubMed] [Google Scholar]
- 4. Eloundou T., Manning S., Mishkin P., and Rock D., “GPTs Are GPTs: Labor Market Impact Potential of LLMs,” Science 384, no. 6702 (2024): 1306–1308. [DOI] [PubMed] [Google Scholar]
- 5. Singla A., Sukharevsky A., Yee L., Chui M., and Hall B., The State of AI: How Organizations Are Rewiring to Capture Value (McKinsey & Company, 2025), https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai.
- 6. Lockhart A., Automation Nation? AI Adoption in Canadian Businesses (Toronto Metropolitan University; The Dais, 2023), https://dais.ca/reports/automation-nation-ai-adoption-in-canadian-businesses/.
- 7. Bengio Y., Mindermann S., Privitera D., et al., “International AI Safety Report,” preprint, arXiv, 2025, arXiv:250117805.
- 8.Goldman Sachs, Generative AI Could Raise Global GDP by 7% (Goldman Sachs, 2023), https://www.goldmansachs.com/insights/articles/generative-ai-could-raise-global-gdp-by-7-percent.
- 9. Zorzenon R., Lizarelli F. L., and de A. Moura D. B. A., “What Is the Potential Impact of Industry 4.0 on Health and Safety at Work?,” Safety Science 153 (2022): 105802. [Google Scholar]
- 10. Fisher E., Flynn M. A., Pratap P., and Vietas J. A., “Occupational Safety and Health Equity Impacts of Artificial Intelligence: A Scoping Review,” International Journal of Environmental Research and Public Health 20, no. 13 (2023): 6221. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Howard J., “Artificial Intelligence: Implications for the Future of Work,” American Journal of Industrial Medicine 62, no. 11 (2019): 917–926. [DOI] [PubMed] [Google Scholar]
- 12. Shah I. A. and Mishra S., “Artificial Intelligence in Advancing Occupational Health and Safety: An Encapsulation of Developments,” Journal of Occupational Health 66, no. 1 (2024): uiad017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Jetha A., Bakhtari H., Irvin E., et al., “Do Occupational Health and Safety Tools That Utilize Artificial Intelligence Have a Measurable Impact on Worker Injury or Illness? Findings From a Systematic Review,” (2024), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4872944. [DOI] [PMC free article] [PubMed]
- 14. Mila ‐ Quebec Artificial Intelligence Institute , Launch of the First International Report on AI Safety Chaired by Yoshua Bengio (Mila, 2025), https://mila.quebec/en/news/launch-of-the-first-international-report-on-ai-safety-chaired-by-yoshua-bengio.
- 15.Leger Marketing, Inc, “Leger Opinion 360 Data Collection Services. Data Collection Powered by Research,” https://leger360.com/leger-opinion/.
- 16.Institute for Work & Health (IWH), “OHS Vulnerability Measure,” https://www.iwh.on.ca/sites/iwh/files/iwh/tools/ohs_vulnerability_measure_guide_2017.pdf.
- 17. Lay A. M., Saunders R., Lifshen M., et al., “The Relationship Between Occupational Health and Safety Vulnerability and Workplace Injury,” Safety Science 94 (2017): 85–93. [Google Scholar]
- 18. Smith P. M., Saunders R., Lifshen M., et al., “The Development of a Conceptual Model and Self‐Reported Measure of Occupational Health and Safety Vulnerability,” Accident Analysis & Prevention 82 (2015): 234–243. [DOI] [PubMed] [Google Scholar]
- 19.“SAS [Computer Program],” Version 9.4. SAS Institute Incorporated. Cary, NC
- 20. Elm E. von, Altman D. G., Egger M., Pocock S. J., Gøtzsche P. C., and Vandenbroucke J. P., “The STROBE Reporting Checklist,” (2025), https://resources.equator-network.org/reporting-guidelines/strobe/strobe-checklist.docx.
- 21. Singla A., Sukharevsky A., Yee L., Chui M., and Hall B., The State of AI in Early 2024: Gen AI Adoption Spikes and Starts to Generate Value (McKinsey & Company, 2024), https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2024.
- 22. Hopkins J. and Bardoel A., “The Future Is Hybrid: How Organisations Are Designing and Supporting Sustainable Hybrid Work Models in Post‐Pandemic Australia,” Sustainability 15, no. 4 (2023): 3086. [Google Scholar]
- 23.Government of British Columbia, Small Business Profile 2023 Highlights (BC Stats, 2023), https://www2.gov.bc.ca/assets/gov/data/statistics/business-industry-trade/small-business/small_business_profile_highlights_2023.pdf.
- 24.Ontario Ministry of Economic Development Job Creation and Trade, “Small Business Sector: Meeting Report,” https://www.ontario.ca/page/small-business-sector-meeting-report#:~:text=Small%20businesses%20are%20crucial%20to,of%20Canada's%20SME%20employer%20businesses.
- 25. Koh D. and Tan A., “Applications and Impact of Industry 4.0: Technological Innovations in Occupational Safety and Health,” Safety and Health at Work 15, no. 4 (2024): 379–381. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplementary Information