Home / Workplace Health & Safety / Can AI be your health and safety manager?

AI at Work | Health and safety management

Can AI be your health and safety manager? We use it every day, and the answer is still NO.

The policy looked perfect. The law it cited doesn't exist.

Author

Monika Csankovszky, Founder

Series

AI at Work

Reading time

About 7 minutes
Let’s talk about the pink elephant, because there’s a question worth asking before another business owner downloads a safety policy and considers the job done.
If AI can write the policy, explain the legislation and produce an inspection checklist, why pay someone to do it? You’ve got customers waiting, employees who need you and another bill on the desk. Being able to ask for a document and have it in seconds is tempting. We understand that.
We also use AI. It helps us edit, research and organize information, and it helped us write this article. Our Academy uses virtual presenters, and we identify them as such. We’re comfortable telling you that.
Our concern begins when a convincing answer becomes the reason someone stops checking.

1. If the answer were wrong, would you recognize it?

That’s the question to ask before handing AI responsibility for your health and safety program.
Our rule is straightforward: use AI for work you can competently check, and get qualified help with what you can’t.
Checking means understanding which requirements apply to your business, how the work is done, what equipment is involved and what could hurt someone. It also means knowing whether the proposed protection will actually work.
With that knowledge, AI can help organize your findings, improve your wording and research something you need to investigate further. You can dictate what you know and let it polish the words. CCOHS recognizes policy drafting as a useful application, with qualified oversight and review.
But if you don’t know how to assess the workplace, asking AI to write your system leaves you with the same knowledge gap and a much more convincing document.

2. You didn't mention the hazard because you didn't know it was a hazard

You tell AI what your company does, how many people you employ and which machines you use. It produces a program that appears to cover everything you described.
Meanwhile, the garbage is rotting beside the back entrance in July. Bob’s tools are scattered across the walkway again. You’ve got five first aid certificates on file, but nobody available on the evening shift. An inspector’s order from last spring still needs attention, and a supervisor’s allowing a shortcut because it gets the job done faster.
Those details weren’t in your description. Some have become so familiar that you’ve stopped noticing them.
“What if I upload pictures?”
Pictures help, as do records and video. Someone still decides what to show, and the camera only captures what’s put in front of it. A photograph of a tidy machine won’t explain how someone clears a blockage when production’s behind.
An assessment needs to investigate how the work actually happens: talking with workers, watching tasks, comparing records and following up when things don’t agree. Some hazards need measurements or specialist advice. A chat conversation alone doesn’t establish that any of this has happened.

3. A wrong answer can arrive wearing very good formatting

Imagine receiving a policy with your company name, numbered sections, responsibilities and a reference to the legislation. It looks ready to print.
Then someone checks the reference, and the section doesn’t exist.
Or the law exists, but it belongs to another province. Perhaps the requirement changed, or the lockout procedure describes equipment you don’t have. The writing can sound just as confident when the information is wrong.
Now consider the harder mistake to catch: the requirement that never made it into the document. How do you question a protection you never knew belonged there?
There’s already a documented warning from a Canadian regulator. On September 10, 2026, WorkSafeNB reported seeing AI-generated safety documents referencing other provinces’ legislation or U.S. standards, containing inaccurate or incomplete information, omitting workplace-specific hazards or required procedures, and creating privacy concerns. Its advice was direct: “Using AI does not replace your duty to exercise due diligence.”
“But I asked it to check its answer.”
That can help, but another reassuring response doesn’t establish that the policy is correct. Before relying on it, ask:
  • Have the requirements been checked against current sources that apply to this business?
  • Does the procedure fit the actual equipment, hazards and work?
  • Have the people responsible been instructed, and has anyone checked whether the procedure works in practice?
Those questions take you beyond how the document reads and into whether it protects someone doing the job.

4. Sometimes the answer you want is the answer you should question

Suppose you explain why your arrangement should be sufficient. You’ve only got a few employees, nobody’s been hurt, and the supervisor’s been doing the work for twenty years. Then you ask AI whether you’re covered.
AI can challenge those assumptions. It can also accept too much of the story you’ve given it, and it’s easy to trust reassurance when it agrees with what you already hoped was true. Overreliance on automated answers is a recognized risk.
A manager or supervisor needs to be willing to say, “I understand why you’ve been doing it that way, but we need to change it.”
Apply that expectation to consultants too, including us. Ask how the conclusion was reached, what was checked and what still needs attention. Paying a person doesn’t make their answer correct automatically.

5. "My AI said so" won't do the employer's job

When a worker needs instruction or a hazard needs attention, someone has to act. Producing a policy doesn’t transfer those responsibilities to the software.
In Ontario, employers must provide information, instruction and supervision, and appoint competent people as supervisors. Competence involves knowledge, training and experience to organize the work and its performance, familiarity with the applicable Act and regulations, and knowledge of actual or potential workplace dangers.
Those responsibilities continue after the document’s saved. Someone needs to check whether the procedure’s being followed, whether workers understand it and whether the protection remains suitable when equipment or work changes.
A policy describing precautions you haven’t taken can become evidence against you. Due diligence depends on the reasonable precautions actually taken before an incident. Records help demonstrate that work; having a policy and a collection of blank forms doesn’t establish that it happened. The same document can fail a prequalification review or a COR audit even when nobody has been hurt.
The same distinction applies to practical competence. A lift truck operator may explain the procedure correctly and still need hands-on instruction and practice. Ontario’s guidance calls for demonstrated operating skills assessed by someone with expert knowledge, with records of what was demonstrated and assessed. A convincing answer in a chat doesn’t demonstrate safe operation in your workplace.

6. That injury report contains someone else's private information

Before uploading it for a summary, look at what you’re sharing.
There may be a worker’s name, medical details and information about their recovery. A harassment complaint may identify several people and contain sensitive allegations.
Before that information goes into an AI service, establish who can access it, how long it’s kept, whether it may be reused and whether you’re authorized to disclose it.
The arrangements differ between services. Canada’s privacy commissioners recommend using anonymized, synthetic or de-identified information where personal information isn’t necessary for the purpose.
Take out identifying details when they aren’t needed, and remember that a name isn’t the only way to identify someone. A job title, location and description of an incident may tell the whole workplace who’s involved.

7. We use AI, and we're still responsible for what we publish

AI helps us prepare drafts, organize material, research questions and communicate more clearly. Those are worthwhile uses, and we expect to keep using it.
We’re responsible for checking what we rely on and what we put in front of our clients.
Our virtual presenters help deliver the Academy’s teaching. Their fluent delivery doesn’t establish that a lesson’s correct any more than a polished document does. The subject matter, the applicable requirements and the instructions still need review.
We tell AI what we know and use it to help us communicate. When it brings us something new, we check it before relying on it. Then we check it again.
That’s the discipline we’re asking business owners to keep too. Use the help, but understand what still needs your attention.

8. Before you train the team, start with yourself

You don’t need to know every answer before you begin learning. You do need somewhere to start, and cost shouldn’t keep you from taking that first step.
The OMHAS Owner Pass gives an eligible owner or director access to every Academy course free, whether the business has employees or the owner works alone. There’s one pass per business, and no payment card is needed. Eligibility terms apply.
It won’t turn you into an auditor or replace qualified help where you need it. It will help you understand responsibilities that might otherwise remain unfamiliar until something goes wrong.
Our courses bring workers, supervisors and owners into the same learning. Everyone learns what they’re responsible for and what they should expect from one another.
Take Lift Trucks. The course connects with the OMHAS lift truck module, including the procedure, inspection checklist and records. It explains what to look for, why it matters and what should happen when something’s wrong. The worker learns what the supervisor’s assessing, and the supervisor learns what to check and how to follow up.
That gives you something useful to discuss before anyone signs the policy. Ask them to explain the procedure and show how they’ll use it.
When you’re ready to bring your team into that learning, the Library Card gives them access to the Academy’s courses. Practical training, workplace-specific instruction and assessment still need to follow where required.
You’re welcome to use AI. Just keep your eyes open, and make sure someone understands the work well enough to check what it gives you.
Explore the Owner Pass and Library Card at omhas.ca/academy.
Have a health and safety question? Bring it to the conversation on our LinkedIn page. Let’s think it through together.
This article opens our AI at Work series: ten posts over two weeks exploring AI’s place in health and safety management and supervision.

Sources

Recent insights

Start here

OMHAS Owner Pass

Every OMHAS Academy course, free for one eligible owner or director.

Need help now?

Talk to an OMHAS advisor about your health and safety program.