AI on the Job · Week 1
The WSIB is pouring money into prediction. The data those predictions run on starts in your workplace.
Two years ago, the WSIB spent roughly $5 million a year on research. Last year it spent $205 million. And a growing share of that money is chasing one idea: spotting injuries before they happen.
That's a big shift. And it has a quiet catch that almost nobody is talking about at the supervisor level.
Welcome to AI on the Job. Every week, we take one real development in artificial intelligence and translate it for the people who actually run Ontario workplaces: supervisors, JHSC members, and small employers. No hype. No vendor pitch. Just what it means on site, how it fits with your duties under the Occupational Health and Safety Act, and one thing you can do this week.
The news
In June 2026, the WSIB committed $63.2 million to create an Institute for Workplace Health at North York General Hospital in Toronto. It will be the GTA's first research centre built to apply AI, real-time predictive modelling and health data to injury prevention and recovery.
Its main target is musculoskeletal injuries: the strains, sprains and overuse injuries that make up about 80 per cent of claims the WSIB receives. The board's CEO described the goal as being able to predict how injuries happen before they happen.
It isn't a one-off. It follows a $65.75 million investment in the Lawson Research Institute in London, which includes data science, AI and virtual reality, and a $20 million centre at Fanshawe College using extended reality and AI to prepare first responders for high-risk situations. The WSIB has also said it intends to keep investing at a similar level over the next several years.
So the direction is clear. Ontario's prevention system is moving from counting injuries to predicting them.
What this means on site
Let's talk about what predictive AI actually runs on. It runs on records. Incident reports. Near-miss reports. First aid logs. Inspection findings. Modified-work notes.
Every one of those starts with a person on your site filling in a form, usually at the end of a long shift.
Here's what that looks like in practice. Two reports describe the same event:
- Report A: "Worker strained back lifting."
- Report B: "Worker felt sharp pain in lower back while lifting a 30 kg pump out of a truck bed alone. Lift was from below knee height, reaching over the tailgate. Second worker was on another task. Pump now moved with a hand truck and two-person rule posted."
Report A tells a model, or a manager, almost nothing. Report B tells you the load, the posture, the staffing gap, and the fix. Multiply that difference across a few hundred reports and you can see why the quality of front-line paperwork is going to matter more than it ever has.
A model can't learn from a near miss nobody wrote down. And it can't learn much from one written in four words.
The OHSA lens
None of this changes who is responsible. Under the OHSA, the employer still has to take every precaution reasonable in the circumstances for the protection of a worker, and supervisors carry the same duty for the workers they direct. A prediction from a piece of software is not a precaution. What you do with it is.
There's a flip side, too. When a tool flags a risk, that flag becomes part of the record. If it sits there and nothing happens, you haven't built a due diligence defence. You've documented what you knew and didn't act on.
AI won't make a weak safety program strong. It will make a weak safety program visible.
That's the real takeaway. The workplaces that benefit from this wave won't be the ones that buy the fanciest platform. They'll be the ones whose basic records were already honest, specific and acted on.
One thing to do this week
Pull your last 10 incident and near-miss reports. For each one, ask four questions:
- Could someone who wasn't there tell what task was being done?
- Does it say what part of the body was affected, or what nearly happened?
- Does it name a cause beyond "worker not paying attention"?
- Does it say what changed afterward?
If fewer than seven of the ten pass, don't lecture the crew. Fix the form. Add prompts for the load, the posture and the conditions, especially for lifting and handling, since that's where most claims come from. Then show supervisors one good example, like Report B above.
The predictive tools are coming either way. Whether they see your workplace clearly is still up to you.
AI on the Job runs every week on WorkSafe Sounds. This post is general information, not legal advice.
Sources
- WSIB commits $63M to North York General workplace health institute (Canadian Occupational Safety, June 11, 2026)
- $65.75M grant positions Lawson as Canadian leader in workplace-injury research (St. Joseph's Health Care London)
- Fanshawe receives $20M from WSIB to establish training lab for first responders (Academica)