The AI assistant wants to do something for me — send, buy, delete. Do I let it?
First, breathe. An assistant that asks for your approval is a system working exactly as it should — that “are you sure?” exists precisely for you. The simple rule: approve only what you would have done yourself, and anything irreversible you press yourself. And if it has already got something wrong, the damage can almost always be limited in the first minutes.
First steps, right now
- Read exactly what it wants to do before you approve: who it is sending to, what amount it is paying, which file it is deleting. Approve only if you would have done exactly the same thing yourself, with your own hands.
- Keep manual approval turned on ("ask me first") for sending, payments and deletions. Don't switch to "auto-approve" to get rid of the questions — the questions are your safety net.
- For shopping and payments, give the assistant as little as possible: a card with a low limit or a virtual card, not your main card and not access to your whole bank account.
- Whatever is irreversible stays with you: permanent deletions, messages to many people, large payments. Let the assistant prepare things (a draft, a shopping basket), but press the final button yourself.
- Has it already done something wrong? Limit the damage now: check the "Sent" folder and recall the message if you can, cancel the order, call your bank about payments (the official number on the back of your card) — and disconnect the assistant from that app until you understand what happened.
What NOT to do
- Don't approve everything it asks out of reflex — the habit of always tapping "Yes" is exactly the risk.
- Don't let it work unsupervised for hours on end with the right to send, buy or delete.
- Don't give it your bank password or your main card — use low limits and virtual cards.
- Don't assume it "knows best" — an assistant can be tricked by an email or a page it reads (with instructions hidden in the text) into doing something other than what you asked.
How to spot it next time
- It proposes an action you didn't ask for: out of nowhere it wants to send an email, open a link or "check" an account.
- It wants to send personal data or files to an address or a website you don't recognise.
- It pushes you to approve quickly or suggests you turn off the confirmations "to be more efficient".
- Its behaviour changes suddenly after it has read a message or a document received from someone else — that text may have contained hidden instructions for the assistant.
This guide is meant to help you act fast. It does not replace official instructions from your bank, the police or the authorities. When in doubt, call the numbers above.