Is AI Safe to Use? A Plain-English Guide to Your Data, Your Money and the Scams to Watch in 2026 (ChatGPT, Gemini, Jobbit)

The most common reason people give for not trying AI is not that it is hard. It is that they do not know what happens to what they type. That is a fair question, and the honest answer is: it depends on the setting, and the setting is usually one click away. Is AI safe to use? Safer than most people fear, less safe than the adverts imply, and almost entirely in your hands once you know five things.

This guide is for the person who has never used an AI tool, or has used one and worried afterwards. It covers what the big tools do with your conversations by default, the seven things you should never paste anywhere, how criminals now use AI against ordinary people, why AI is sometimes confidently wrong, what your rights are in the UK, and a ten-point checklist for using an AI agent for real jobs without handing over more than you meant to.

The short answer

Typing a letter into an AI tool is about as risky as emailing it to a company: the company can read it, keep it and, unless you tell it not to, learn from it. The risk is not the AI itself. The risk is putting something in that you would not put in an email to a stranger, trusting an answer you did not check, or letting a tool act on your money without you looking. Manage those three and the rest is convenience.

What happens to what you type

Most consumer AI tools use your conversations to improve their models unless you switch that off. The table shows the situation for the tools people ask about most, as published in their own help pages in 2026.

Two things matter more than the table. First, once a conversation has been used in a training run it cannot be pulled back out, so change the setting before you start, not after. Second, all of these services keep logs for a period for safety and legal reasons even when training is off. Treat that as normal, in the way your bank keeps records, and decide what you paste accordingly.

The seven things never to paste

1. Passwords, PINs and one-time codes.
2. Full card numbers and the three digits on the back.
3. Your National Insurance number alongside your name and date of birth.
4. Other people's personal details without their permission, especially health information.
5. Anything covered by a confidentiality agreement at work.
6. Photos of documents you would not post publicly, unless the tool keeps them private and you have read how.
7. Login details "so it can do it for me". A proper agent asks you to log in yourself and shows you the screen.

Everything else, your own letters, receipts, plans and questions, is what these tools are for. The line is the same one you use for email.

Can AI take your money?

A chatbot cannot. It writes text. An AI agent that browses and acts, which ChatGPT, Gemini, Copilot, Claude, Manus and Meta's Muse now offer in some form, can fill in a form and, if you have given it a card or a logged-in session, buy something. That is the feature, and it is also the thing to supervise. Three rules keep it safe:

• Keep spending behind your approval. Choose tools that stop and ask before they pay, submit or delete, and set them to plan first for anything that changes money.
• Watch it work. Jobbit's agent runs in a browser you can watch and take control of; when a website asks for a human, you step in, then hand back.
• Give it the least it needs. Files and passwords for the job, not for your life. Jobbit's Projects keep each job's files and secrets separate for this reason.

The wider web is tightening too: since 15 September 2026 the largest content network blocks AI agents by default on many sites unless the agent proves who it is, after a summer in which several companies' agents were found reaching pages they should not have. Expect an agent to meet more "prove you are human" walls, and treat that as the system working.

How criminals use AI against you

This is the part that deserves the most care, because it does not involve you using AI at all. Three seconds of a voice is enough to clone it, according to McAfee's research, and cloned-voice phone scams have grown faster than any other kind. Deepfake video adverts featuring well-known faces sold a fake investment that took £27 million from more than 6,000 people in the UK and Canada in one 2024 case. Complaints about fraud and scams to the UK Financial Ombudsman rose by around 80% between 2023 and mid-2025.

• A family voice asking for money urgently. Hang up and call the person back on the number you already have. Agree a family safe word this week; it costs nothing.
• A celebrity or a bank in a video telling you to invest or move money. No genuine bank asks you to move money to a "safe account". Ever.
• A message that knows a lot about you. AI makes it cheap to write a convincing email to one person. Check the sender's address, and never click a link to log in; type the address yourself.
• Anyone rushing you. Urgency is the tell. Real organisations survive you calling back tomorrow.

When AI is confidently wrong

AI tools sometimes state things that are not true in the same confident tone they use for things that are. The technical word is hallucination; the practical meaning is "check the specifics". Names, dates, prices, opening hours, legal thresholds and medical doses are the usual slips. The habit that fixes it: ask "where did that come from?" and expect a page you can open, and read anything you are about to send, sign or pay for. Use AI to understand a letter from the hospital or HMRC and to prepare your questions; ask a person who is responsible for the answer before you act on health, law or tax. The beginners' guide on this site lists twelve everyday jobs where this habit is all you need.

Your rights in the UK

UK data protection law applies to AI companies like any other. You have the right to ask what a company holds about you, to have it corrected or deleted where the law allows, and to complain to the Information Commissioner's Office if a company will not answer. Companies that serve UK customers must explain in plain terms what they do with your data, and Ofcom's online safety rules now cover AI chatbots that can generate harmful content. None of this stops you being careful; it does mean the law is on your side when you ask questions.

The ten-point checklist

1. Turn off model training in the settings of any tool you use before your first real conversation.
2. Use a temporary or private mode for anything you do not want kept.
3. Never paste the seven things above.
4. Log in to websites yourself; do not hand an agent your passwords.
5. Keep paying, submitting and deleting behind your own approval.
6. Use a tool that shows you its browser and its working.
7. Give each job its own project with only the files it needs.
8. Check names, dates, prices and deadlines before you act.
9. Agree a family safe word and never move money on a phone call.
10. Read the privacy terms once, including ours, and ask the company if something is unclear.

Why Jobbit is built around supervision

We built the agent for people who want to see what it is doing. It works in a browser you can watch and take control of, it can be set to plan before it acts, each project keeps its own files and secrets, and when a job needs a person, a vetted professional from the human network takes it with payment held in escrow until you accept the work. The Free plan is 100 credits a day with no card, which is enough to try everything in this guide before you decide anything.

Frequently asked questions

Is it safe to give an AI my personal documents?

For your own admin, yes, with the training setting off and the seven items above kept out. Read how the tool stores files, use a private mode for anything sensitive, and prefer tools that keep each job's files separate.

Does ChatGPT use my conversations to train its models?

On the Free and Plus plans, yes by default; you can turn it off under Data Controls, and Temporary Chats are never used. Pro, Enterprise and Edu plans are excluded. Conversations already used cannot be removed afterwards.

Can an AI agent spend my money?

Only if you give it a card or a logged-in session and let it act unsupervised. Keep payments behind your approval, watch the browser, and give it only the access the job needs.

How do I spot an AI scam?

Urgency, a request to move money, a familiar voice or face asking for something unusual, and a link to log in. Hang up, call back on a known number, type website addresses yourself, and agree a family safe word.

What if the AI gives me wrong information?

Ask where it came from and open the page. Check specifics before you send or pay, and confirm anything about health, law or tax with a responsible professional.

How do I try an AI agent safely for the first time?

Start with a job that involves your own documents and no money, such as a letter or a comparison, with training off and a private mode on. Jobbit's agent shows you its browser, plans first when you ask, and costs nothing to try on the Free plan's 100 credits a day.