AI safety isn’t a policy document. It’s how your systems actually behave.

Ask an owner what “AI safety” means and most point at a vendor’s compliance page. That’s not what actually protects your business. Safety is a design decision made once, at the start, in how a system is built — not a document you read afterward.

The real questions are specific. Where does the data live, and who can see it? Does anything you feed in get used to train someone else’s model? What happens the moment something goes wrong — does a person see it, or does it just keep running?

That last one matters most. Every AI team member we build has a human checkpoint on anything that leaves your business or costs you money — a draft goes out for review, a reply sits in a queue, an action needs a yes. The AI does the work; a person still owns the outcome.

The second thing worth checking is control, not just access. Your data should sit in your own accounts — hosted in Australia, under your control — not locked inside a vendor’s platform you’d have to rebuild from scratch to leave. If a tool can’t explain simply where your information goes, that’s the answer.

None of this is exotic. It’s the same due diligence you’d apply to any new hire or supplier — just applied to software that now sits in the middle of client work. Ask the boring questions before you ask the exciting ones.