Lockout/tagout steps, hazard assessments, and toolbox talks used to come from a safety manager’s desk. More and more, a generative AI tool writes the first draft, and that draft doesn't always get checked before it reaches the floor. When one of those documents gets a step wrong, the people who follow it want to know who answers for it.
Insurance Business reported that insurers have filed 4,078 state-level form adoptions of ISO's generative AI exclusions across 49 states and DC through July 31, covering several commercial insurance lines. That raises the possibility that businesses relying on AI could face gaps in liability coverage when something goes wrong.
TRADESAFE surveyed 514 U.S. workers in manufacturing, construction, warehousing, logistics, and the skilled trades about how AI is being used in safety documentation. We asked who reviews AI-drafted material, how much workers trust it, and who they’d hold accountable when an AI-assisted error leads to a near miss or an injury.
Our research shows the workforce side of that question. Among respondents with visibility into how safety documents are created, 51% say their company uses generative AI to help write them, yet only 18% of respondents report a formal policy at their workplace defining who is accountable when AI-assisted safety content contains an error.
Key Takeaways
- Among workers with visibility into document creation, 51% say their company uses generative AI to help write safety documents.
- Only 18% of workers report a formal policy defining who is accountable when AI-assisted safety content contains an error.
- One in five workers (20%) report having seen an adverse outcome associated with an AI safety-document error, including near misses, minor injuries, and failed inspections.
- In the unreviewed-AI injury scenario, 56% look to company leadership or the safety department (28% each), versus 9% who blame the employee handling the document.
- More than a third of workers (36%) have personally seen an AI-generated safety document with an outdated requirement, vague instruction, or missing hazard step.
- Nearly two-thirds of workers (63%) would trust a safety policy less if they knew it was partially AI-written, while 78% say qualified human review increases trust.
- Frontline workers are 45 percentage points less likely than leadership to say they have visibility into how safety documents are created (31% vs. 76%).
AI Is Already Writing Your Safety Rules
Generative AI has moved into safety paperwork faster than many workplaces have built checks around it.
AI use and oversight varied widely across the companies we surveyed:
- Among workers with visibility into how safety documents are created, 51% say their organization already uses generative AI for safety-related documents.
- Only 49% of workers say those AI outputs are independently reviewed before use, and the answer depends on who you ask. Company leadership (74%) is far more likely to report independent review than supervisors or managers (52%), frontline workers (36%), or safety professionals (23%).
- Frontline workers are 45 percentage points less likely than leadership to say they have visibility into how safety documents are created (31% vs. 76%).
- More than one-third of workers (35%) say the employee who creates the document is responsible for approving it, while only 2% say no one formally signs off.
- Just over a third of workers (34%) say AI is being adopted for safety work faster than their company can properly oversee it.
- Nearly half (47%) of workers think an uncaught AI error could happen at their own organization, and over a third (36%) have personally seen an AI-generated safety document containing an error.
Trust takes a hit once AI enters the picture, though a human reviewer goes a long way toward earning it back:
- Almost two-thirds of workers (63%) say they trust a safety policy less if they know it is partially AI-written.
- That skepticism runs higher on the floor. Frontline workers (67%) are more likely than managers (61%) to trust a policy less if they know AI wrote it.
- Nearly four-fifths of workers (78%) say a safety document earns more trust when a qualified person has reviewed it, whether or not AI helps write it.
When a Safety Document Goes Wrong, Workers Look Up the Chain
We gave every worker the same story about a safety procedure with a mistake in it, and changed only how that procedure was created.
Each worker saw just one of three versions of the scenario:
-
Written by a person (33% of workers). An employee wrote the procedure, and the mistake was in it.
-
AI draft, checked by a person (32% of workers). AI drafted the procedure, and a person reviewed it before use.
-
AI draft, never checked (35% of workers). AI drafted the procedure, and it went straight to the floor with no review.
We then asked who was most responsible when the mistake was caught in time and when it caused a serious injury.
Skipping the review moved blame away from the individual and toward the people running the operation:
- After a serious injury, workers were most likely to name company leadership or the safety department when no one had checked the AI draft (56%, or 28% each).
- That share was lower for the procedure written by a person (49%, with 24% naming leadership and 25% the safety department) and for the AI draft that was checked (39%, with 19% and 20%).
- Blame on the employee who created or handled the procedure moved the other way. It came in at 17% for the person-written version and 19% for the checked AI draft, but just 9% for the unchecked one.
- The same pattern held when the mistake was caught in time. Blame on the employee fell from 30% for the person-written procedure to 24% for the checked AI draft and 14% for the unchecked one.
Looking at all three versions together, workers spread the blame fairly widely when the mistake was caught before anyone got hurt:
- The safety department is the most-blamed party, named by 27% of workers across all scenarios.
- The employee who handles the document follows (23%), then the supervisor or manager (14%) and company leadership (13%).
- Few point at the worker who follows the procedure (6%) or the company that made the AI tool (5%), while 13% say responsibility should be shared.
- Workers are more than four times as likely to blame the employee (23%) as the AI provider (5%).
Across all scenarios, an injury moves accountability toward the top of the organization:
- Across all scenarios, blame for company leadership nearly doubles, from 13% when the error is caught to 24% when it causes an injury.
- The employee’s share falls from 23% to 15% over that same shift.
- When an unreviewed AI error leads to an injury, 56% of workers blame company leadership or the safety department (28% each), compared with 9% who blame the employee.
- Close to two-thirds (64%) agree that blaming AI for a mistake usually means the company’s oversight isn’t clear enough, yet only 18% of organizations have a formal policy defining accountability for AI-assisted safety content.
For some workers, the risk has already moved past the hypothetical:
- One in five workers (20%) have already seen an adverse outcome tied to an AI safety-document error. Near misses are most common (11%), followed by minor injuries (7%), failed inspections (6%), equipment or property damage (5%), and serious injuries (3%). Respondents could select more than one outcome.
- More than two-fifths (44%) agree that using AI to write safety documents puts workers at greater risk.
- Most workers wouldn’t blindly trust the tool. If an AI instruction conflicted with what they knew was safe, 48% would ask a supervisor, 37% would stop and challenge it, 10% would use their own judgment, and only 4% would follow it as written.
Expert Commentary
What did you find most interesting in this research?
"What stood out is where workers place responsibility when an unreviewed AI-drafted safety document leads to an injury in the survey scenario. They look primarily to company leadership and the safety department. That raises an important question for employers: as AI becomes part of safety documentation, are review and approval responsibilities clear to the people creating and using those documents?" — Herbert Post, Vice President, TRADESAFE
What advice do you have based on the findings?
"Organizations should treat AI-drafted safety content as a first draft that requires named sign-off from a qualified person, and document who performed that review. Accountability needs to be defined before an incident. Employers should also explain the review process to their workforce: nearly two-thirds of respondents say they trust a policy less when AI helped write it, while 78% say qualified human review increases trust." — Herbert Post, Vice President, TRADESAFE
Methodology
TRADESAFE surveyed 514 U.S. adults ages 18 and older working in manufacturing, construction, warehousing, logistics, and the skilled trades about the use of generative AI in workplace safety documentation. Respondents were recruited through CloudResearch Connect and Prolific in September 2026.
The survey included a randomized scenario experiment in which each respondent saw one of three versions of a safety-error vignette (human-written, AI-drafted and then reviewed, or AI-drafted with no review), with version A (n=169), version B (n=167), and version C (n=178). Several questions used skip logic, so some findings are based on smaller groups. For example, AI tool and document questions were shown only to the 202 respondents whose organizations use AI for safety materials, and the review-and-approval question was shown to a base of 317. The AI-use question was shown to the 399 respondents with at least some visibility into how safety documents are created.
The sample was 59% male and 40% female. By generation, respondents were 50% millennials, 23% Gen X, 22% Gen Z, and 4% baby boomers or older. By role, 42% were frontline workers, 39% were supervisors or managers, 10% were company leadership, and 9% were safety professionals. The largest industries were manufacturing (33%), construction (19%), logistics (12%), warehousing (11%), and skilled trades (10%). Generations were defined as Gen Z (18 to 29), millennials (30 to 45), Gen X (46 to 61), and baby boomers (62 and older).
One response with an invalid age entry was excluded from generational analysis. All percentages are rounded to whole numbers, and any that do not total 100% are due to rounding. Select-all-that-apply questions are calculated on the full applicable base, so totals can exceed 100%.
About TRADESAFE
TRADESAFE provides industry-leading safety solutions — including Lockout Tagout devices, safety showers, eye wash stations, spill containment solutions, and workplace safety signs — precision-engineered for durability, compliance, and seamless integration into industrial environments. Designed to exceed OSHA, ANSI, and EPA standards, our solutions are relied upon by the nation's top companies, municipalities, and government agencies.
Fair Use Statement
You’re welcome to share these findings for noncommercial purposes. Please credit TRADESAFE and link back to this page so your readers can see the full study.
