Using AI tools safely on site

This talk is about the chat tools and the texts that now turn up on site. It is not a health-and-safety regulation. The hazards above, height, dust, asbestos, vibration, and the floor, still come from HSE. This one is about not making those jobs worse with a phone.

A chat tool can put notes under headings. It has not walked this job. The guide on risk assessments and method statements says what that means in practice. You write what you saw: the address, who else is around, the access, the dust, the cables, the rot, the kit you will actually bring. Then you can ask the tool to sort those notes. You tell it not to add a control you did not mention. If it writes a harness, a permit, or a qualification you do not have, that line comes out. You read the draft against your notes before anyone treats it as the method. A tidy PDF is not a RAMS. The person responsible is still the person who decided the method.

Do not paste a customer's personal data into a chatbot to make the paragraph better. The same guide says not to paste a customer's health details. 'The occupier is frail and cannot leave the room' can be written as 'occupier will remain in the house; keep a clear route and control dust', without the medical story. Do not paste a Unique Taxpayer Reference, a National Insurance number, or bank details. The CIS guide says a job name and the figures you can check are enough. Do not paste a one-time code from a bank text. The scam-text guide says delete the code before you paste the message, if you use a prompt to slow yourself down. A code is a key.

Scam texts now read like a person. Bad spelling is a weak clue. The guide for trades lists the usual ones: a fake toll or van charge, a parcel you were not expecting, an HMRC refund, a supplier who has 'changed bank details', a customer who wants a deposit before you have seen the job, and your bank asking you to read out a code. You were not in the middle of that thing. That is the clue.

Do not tap the link. Do not reply. Do not call the number in the message. If a supplier's account has changed, phone them on the number you already had, not the number in the text. Customer payments go to the account on the invoice you sent. If a customer says you sent new details, they check with you in the thread you already use.

Forward a suspicious text to 7726. That report is free. Note the number it came from, forward the message, and reply with the sender's number when you are asked. 7726 is for the message. It is not where you report that money has already left the account. Phone the bank on the number on the card, or in the app you opened yourself. Then, in England, Wales, and Northern Ireland, report it to Report Fraud or call 0300 123 2040. In Scotland, Police Scotland on 101. Those are the routes in the guide, taken from Report Fraud and the National Cyber Security Centre. Do not pay a 'recovery' service that texts you next.

An apprentice with the work phone needs this in one minute, not a lecture. No links in texts about tolls, parcels, HMRC, the bank, or a customer you do not know. No reading out a login code. No new supplier bank details from a text. If money has gone, the bank first, then Report Fraud or Police Scotland as above.

If someone has already pasted a customer's details, or a code, into a chat, say so now. We can change the password from a device we trust, and we can tell the customer if their information went somewhere it should not. Pretending it did not happen is how a small mistake becomes a bigger one.