AI in Workplace Safety Moves Beyond Experimentation
New research reveals gaps in AI adoption across industries as safety leaders work to turn emerging technology into practical tools.
- By Stasia DeMarco
- Nov 11, 2026
Artificial intelligence is becoming a familiar part of workplace safety programs, but how widely organizations use the technology varies significantly across industries. While some companies are integrating AI into multiple safety functions, others are still testing individual applications or struggling to move beyond basic automation.
New research from Wolters Kluwer Enablon and the National Safety Council (NSC) suggests that the conversation surrounding AI in environment, health and safety (EHS) is shifting from whether organizations should adopt the technology to how they can make it effective in everyday operations.
The research, detailed in The Safety Shift: EHS Readiness in 2026 and a subsequent industry analysis titled Evaluating AI Adoption for EHS Across U.S. Industries, surveyed 1,053 U.S. safety leaders and professionals. Read the full report, Evaluating AI Adoption for EHS Across U.S. Industries (PDF)
Among respondents, 20% reported extensive AI integration across their EHS programs, while 62% described moderate or limited use. Another 16% were exploring or piloting AI technology, and only 2% reported no current use or plans.
Despite those adoption figures, implementation remains uneven. Some industries have developed the digital infrastructure needed to support more advanced applications, while others still face challenges with data quality, technical resources, and organizational processes.
For safety managers, understanding those differences may be as important as understanding the technology itself.
Industry Adoption Reveals Significant Differences
The study identified energy and chemicals, information technology, oil and gas, and manufacturing as leaders in AI adoption for EHS.
Among energy and chemical industry respondents, 76% reported moderate or extensive AI use. Oil and gas followed at 71%, information technology at 70%, and manufacturing at 68%.
Other industries reported lower adoption levels. Transportation, logistics and warehousing registered 62%, followed by food and beverage at 55%, construction and engineering at 51%, pharmaceuticals and biotechnology at 46%, and mining and metals at 39%.
The results suggest that industries with established digital systems and substantial operational risks may be better positioned to incorporate AI into existing safety processes.
Raj Jayaraman, who leads product management and product strategy for Wolters Kluwer Enablon, said the differences largely reflect investments organizations made before AI became widely available.
"If you think of the industries which are further ahead, like oil and gas and manufacturing," Jayaraman said in an interview with Occupational Health & Safety, "they have already been investing in automation. They already have sensors in place. They have connected equipment. They have digital operating systems."
Those investments provide a foundation of information that AI systems can analyze and use.
Construction presents different challenges. Work moves between job sites, contractors, and crews, and important safety information may be recorded in separate systems or on paper. Mining operations can face similar obstacles because of remote locations and complex working environments.
"Fundamentally, AI is only as useful as the information you can give it," Jayaraman said.
For organizations operating without integrated digital records, moving directly into advanced AI applications may prove difficult.
From Administrative Tasks to Safety Insights
Although AI is often discussed as a tool for predicting accidents and preventing injuries, Jayaraman said many current applications remain focused on administrative work and information management.
One example is incident reporting.
A worker may submit a brief description of an incident, leaving the safety manager to verify details, complete forms and review previous events. AI can help organize the initial information, identify missing elements and populate portions of an incident report.
Technology can also help draft descriptions, summarize investigation materials, and identify recurring patterns across workplace records.
Rather than manually searching through inspection reports, procedures and incident histories, a safety professional could use AI to identify repeated hazards or determine which types of incidents are increasing at a particular location.
These applications may reduce time spent entering information and reviewing records, but Jayaraman cautioned that most organizations have not yet connected individual AI tools into comprehensive safety management processes.
"They're often built using existing tools. They're really not part of an end-to-end workflow within an organization," he said.
That distinction matters because automating documentation does not necessarily translate into fewer workplace injuries.
The greater opportunity, according to Jayaraman, is to let safety professionals spend less time on administrative responsibilities and more time working directly with employees, evaluating hazards, and implementing preventive measures.
Why Data Quality Matters
Expanding AI across safety operations requires more than purchasing software.
Organizations need reliable data, consistent reporting practices and a clear understanding of where technology can provide measurable benefits.
A company may maintain incident records in one system, inspection findings in another and corrective-action documentation in spreadsheets. When information is incomplete or disconnected, AI may struggle to produce dependable results.
Jayaraman warned that adding AI to poorly organized information can create additional problems rather than solve existing ones.
The research also identified concerns about the quality and transparency of AI-generated information.
During the interview, Jayaraman cited survey findings showing that 42% of respondents raised poor data quality as a concern, while 38% expressed concerns about a lack of transparency in how AI makes decisions.
Those findings reinforce the importance of validating the information AI systems use, particularly when recommendations could affect high-risk work.
For safety professionals, the issue extends beyond whether a system can produce an answer. They must also be able to determine whether the answer is accurate, relevant to actual operating conditions and supported by reliable information.
Human Oversight Remains Essential
One of the biggest concerns about AI in workplace safety is the possibility of inaccurate or inconsistent recommendations.
Unlike traditional software that follows predetermined instructions, generative AI can produce different responses to similar questions.
In a safety-critical environment, those inconsistencies could have serious consequences if organizations rely on AI-generated recommendations without appropriate review.
"AI should support the decision, not own the decision," Jayaraman said.
He emphasized that safety professionals must remain accountable for final decisions and apply their knowledge of equipment, work conditions and regulatory requirements.
A work permit provides one example.
AI may review information about a proposed job, identify potential hazards and flag unusual conditions. However, an authorized employee should remain responsible for determining whether work can proceed.
Organizations also need clearly defined boundaries governing AI use, including which tasks can be automated, when human review is required and who has authority to approve safety-related actions.
Organizations should establish these safeguards before deploying technology, not after problems arise.
For companies using AI to assist with incident investigations, hazard assessments or compliance activities, maintaining a reviewable record of the information and decisions involved is another important consideration.
Starting Small and Measuring Results
For organizations with limited budgets or technical support, adopting AI does not necessarily require a comprehensive technology overhaul.
Jayaraman recommended beginning with one specific operational problem rather than attempting to introduce AI across an entire safety program.
An organization might start by examining the amount of time employees spend preparing incident reports, reviewing inspection records or searching for information across multiple systems.
Before purchasing additional technology, safety managers should determine whether existing software already includes AI capabilities that could address those needs.
They should also evaluate the quality of available data and establish a way to measure whether the application produces meaningful improvements.
Potential measures include reduced administrative time, more complete incident documentation or faster identification of recurring safety concerns.
Jayaraman also recommended establishing clear responsibilities and review procedures from the beginning.
The objective is to identify a manageable application, evaluate its effectiveness and expand only when the organization can demonstrate value.
That approach may be particularly relevant in construction and mining, where survey respondents reported lower levels of AI adoption despite generally favorable attitudes toward the technology.
Moving Toward More Proactive Safety
Looking ahead, Jayaraman anticipates three changes in the role of AI within workplace safety programs.
The first involves moving beyond isolated tasks toward integrated workflows that connect incident reporting, investigations, corrective actions and follow-up activities.
The second is a gradual shift from reactive applications to more proactive risk identification.
Rather than reviewing information only after an accident occurs, AI could potentially analyze misses, inspections, observations and other records to identify emerging warning signs.
The third change involves the safety professional's role.
As AI takes on more routine administrative work, safety managers may have more time to focus on field activities, employee communication, and decisions involving serious hazards.
Those changes remain potential developments rather than established outcomes. The survey primarily measures reported technology adoption and organizational readiness, not whether AI has demonstrably reduced injuries or fatalities.
For organizations considering their next steps, the findings suggest that practical implementation, data management and human accountability will be central to determining how effectively AI supports workplace safety.
The challenge is no longer simply introducing AI into EHS operations. It ensures that the technology produces reliable information, supports informed decisions and ultimately contributes to safer working conditions.