Survey Finds Errors in AI-Generated Safety Procedures
Workers report mistakes in AI-assisted safety documents, raising concerns about oversight and accountability.
- By Stasia DeMarco
- Nov 06, 2026
More than one-third of workers surveyed reported errors in AI-generated workplace safety documents, while 20% said they witnessed an adverse outcome linked to those mistakes, according to a study by TRADESAFE.
The study, published Oct. 6, surveyed 514 U.S. workers in manufacturing, construction, warehousing, logistics and skilled trades about the use of artificial intelligence in workplace safety documentation.
Among respondents, 36% reported seeing AI-generated documents containing outdated requirements, vague instructions or missing hazard information. Reported adverse outcomes included near misses, injuries, failed inspections and equipment damage.
“Safety procedures can look complete on the surface but still lack steps to protect people from serious injury,” said Herbert Post, vice president of TRADESAFE. “That's what makes errors from AI particularly concerning.”
The research also found that 51% of respondents familiar with how their organizations create safety documents said their employers use generative AI to help write them.
However, independent review practices varied by job role. Among company leaders surveyed, 74% reported that AI-generated safety materials undergo independent review, compared with 52% of supervisors and managers, 36% of frontline workers and 23% of safety professionals.
Post said AI can help draft safety procedures but cannot independently evaluate every hazard associated with equipment or working conditions.
“Technically reviewing the document is where missing steps, outdated requirements, and instructions that are not congruent with the work should be identified,” Post said.
Questions About Accountability
The survey also found that only 18% of respondents reported having a formal workplace policy defining accountability for errors in AI-generated safety documents.
In a hypothetical scenario involving a serious injury caused by an unreviewed AI-generated procedure, 56% of respondents assigned primary responsibility to company leadership or the safety department.
Worker confidence in AI-assisted safety documents also varied. Nearly two-thirds of respondents (63%) said they would trust a safety policy less if they knew AI had helped write it. However, 78% said a qualified person's review would increase their confidence.
“Employers should have systems in place for workers to identify potentially unsafe situations, stop work, and have clear guidance regarding who reviews and approves documents assisted by AI,” Post said.
The survey was conducted in September 2026 using respondents recruited through CloudResearch Connect and Prolific. Some findings were based on smaller groups because of screening questions.
The results reflect workers' reported experiences and perceptions, not independently verified workplace incidents.
About the Author
Stasia DeMarco is the content editor of Occupational Health & Safety.