Safety Leaders Build A.I. Programs Slowly as Ethics and Data Gaps Persist
Only 5% of organizations consider their environmental health and safety programs fully integrated with artificial intelligence, while 58% remain in discovery or piloting stages, according to a survey released by the American Society of Safety Professionals.
Two-thirds of organizations began their A.I.-EHS journey within the last 18 months, and most expect to take one to two years to move from early experimentation toward scaling and integration.
The findings come from an online survey of nearly 1,000 safety professionals conducted from July 6 through Aug. 2, 2026. ASSP’s research identified five stages of A.I.-EHS adoption: non-adopters, discovery, piloting, scaling and integrated. While most organizations remain in the exploratory stages of discovery and piloting today, more than half expect to be in the transformative stages of scaling and integrated within two years, and the share expecting to be fully integrated rises from 1 in 20 today to 1 in 5 in two years.
Adoption varies significantly by organization size and industry: 45% of organizations with 5,000 or more employees are classified as transformative, compared with 19% of those with fewer than 250 employees.
Among industries, insurance leads adoption, with half of insurance organizations reaching transformative status, compared with a third or fewer in every other industry surveyed.
Early Gains Are Easier Than Advanced Outcomes
Organizations do not need full A.I. integration to see results, ASSP found. Reducing time spent on repetitive tasks and generating more actionable insights from data emerge as early wins, cited as goals by 75% and 64%, respectively, of organizations in the discovery and piloting stages. Report development and tailored education and training are among the most effective entry-level applications, rated effective by 79% and 68% of users, respectively.
More advanced capabilities take longer to materialize. Predicting risk with greater accuracy and assessing or responding to risk in real time remain limited until organizations reach full integration. Real-time monitoring and alerts carry the lowest effectiveness rating of any application studied, at 49%.
Cost savings and reductions in serious injury and fatality rates remain largely out of reach even for integrated organizations, which ASSP attributed partly to upfront investment costs outpacing returns, and to safety professionals expressing far more confidence in A.I.’s ability to organize information and detect patterns than in its ability to identify or mitigate SIF-related risks.
Ethical Concerns Persist Even as Experience Grows
Experience with A.I. tempers but does not eliminate ethical concerns, according to ASSP. About 5 in 10 non-adopters expressed a great deal of concern about the ethical questions and risks associated with A.I. in EHS, compared with about 3 in 10 professionals at fully integrated organizations.
Concern runs especially high around human oversight: 80% expressed concern about automatically generated actions such as discipline or risk-cause assignments made with little or no human validation, including 48% who expressed a great deal of concern. Concern was even more widespread, at 85%, about decisions made from inaccurate information or A.I. hallucinations.
Governance has not kept pace with adoption, ASSP found. While 62% of integrated organizations have A.I.-EHS ethics policies or guidelines, compared with just 20% of discovery-stage organizations, nearly half of all organizations surveyed still lack any A.I. ethics policies.
One discovery-stage respondent from the insurance industry said in the survey: “Validation needs to be part of the process to assure inputs are being utilized correctly and outputs are real. No hallucinations.”
Data Quality and Training Gaps Remain Widespread
Data quality is a barrier at every stage of A.I. adoption, cited by 62% of organizations overall, according to ASSP. Only 33% of discovery-stage organizations rated their data quality as good or excellent, compared with 56% at the scaling stage and 70% at full integration. Biometric data collection remains rare, with just 8% of organizations currently gathering it.
Training gaps are also widespread. About 2 in 5 EHS professionals have never attended formal A.I. training, and even among transformative organizations, 1 in 4 report no formal training, ASSP found. Only 32% of safety professionals at integrated organizations have received EHS-specific A.I. training. Professionals who have attended training report notably higher A.I.-EHS knowledge (75% vs. 56%) and comfort interpreting A.I.-generated metrics (79% vs. 61%) compared with those without training.
Obtain the full survey here. &

