A big safety organisation, the National Safety Council, recently asked more than a thousand safety and operations people what they thought about AI. The headline everyone picked up was that AI is now common in safety work. Good news, you'd think.

But there was a more interesting number underneath. Nine in ten of those same people said they had worries about AI. And around two thirds said their main worry was people trusting it too much.

I think that's a healthy sign.

Here's why. When a new tool turns up in a high-risk job, the worst thing you can do is just go along with it. The best people ask hard questions first. What happens when it goes wrong? Who is to blame when it does? Will the people on the ground actually trust it enough to use it? Those aren't the questions of someone being difficult. They're the questions of someone who has seen things go wrong and wants to stop it happening again.

AI in safety is useful. I wouldn't be building Rivisio if I didn't believe that. But useful doesn't mean always right, and it doesn't mean you switch your brain off.

The worry about trusting it too much is a fair one. Say an AI tool tells a safety manager a site is low risk, and because of that the manager stops walking the floor or stops talking to the crew. The tool hasn't made anyone safer. It has just given false confidence. That's worse than having no tool at all.

The way I see it, AI should make good safety people better at their job, not replace the judgement they've built up over years. A good tool spots patterns in thousands of records that a person would never have time to find. It points out the things that need attention before they turn into accidents. It takes the paperwork off your plate so you can spend more time on site, not less.

But it should never have the final word. A person who knows the site, the crew and the job still needs to be the one who decides.

The rules around this are starting to catch up. More firms are writing simple policies that say who is responsible when an AI tool makes a call, that keep a clear record of how it reached its answer, and that check its results against what is really happening on site instead of just taking its word for it.

That's the right way to do it. It isn't anti-AI. It's how you build something people can actually trust.

If your crew is unsure about AI, don't brush that off. Use it. Ask them what they would need to see to feel comfortable with it. Start there. Because a tool nobody trusts won't change a thing, however clever it is underneath.