Is AI safe? What to share, what not to, and how to check

It’s sensible to ask. The good news: used sensibly, AI tools like ChatGPT and Claude are about as safe as other everyday online services like email or online banking. Here’s what “sensibly” means.

The traffic-light rule

🟢 Fine to share

• General questions and everyday tasks
• Drafts of emails, letters and documents (with private details removed)
• Your preferences and general situation (“I’m a teacher”, “I live in Bristol”)

🟠 Think first

• Work documents: check your employer’s policy. Many workplaces have rules about which AI tools you can use.
• Other people’s personal information: remove names and details if you can
• Health information: fine for general questions, but think about how specific you need to be

🔴 Never share

• Passwords, PINs or security codes
• Full bank card or account numbers
• National Insurance or passport numbers
• Anything you’d be devastated to see made public

Five settings to check today

1. Training. Both tools let you choose whether your chats can be used to improve future versions. In ChatGPT: Settings → Data controls. In Claude: Settings → Privacy. Many people turn this off.
2. Memory. Decide whether you want the AI to remember things between chats. You can review and delete what it remembers.
3. Private chats. Use Temporary chat (ChatGPT) or incognito (Claude) for one-off conversations you don’t want saved.
4. Two-step verification. Turn it on for your AI account, or for the email or Google account you sign in with.
5. Chat history. You can delete individual conversations, or all of them, at any time.

AI can be confidently wrong

AI sometimes makes things up: a fact, a date, a quote, a website that doesn’t exist. This is sometimes called a “hallucination”. It usually sounds just as confident as when it’s right.

So for anything that matters:

• Ask for sources, then check them yourself
• Cross-check important facts with a trusted website (NHS, GOV.UK, Citizens Advice)
• Never rely on it alone for medical, legal or financial decisions

Our guide When AI gets it wrong goes deeper.

Watch out for AI-powered scams

Scammers use AI too. Be extra wary of:

• Fake apps. Download ChatGPT only from OpenAI and Claude only from Anthropic, via the official app stores.
• Voice clones. A call from a “family member” in trouble asking for money? Hang up and call them back on a number you know. Some families agree a secret code word.
• Too-perfect messages. Scam emails no longer have obvious spelling mistakes. Judge the request, not the writing: urgency plus money plus secrecy is a red flag.
• “AI investment” schemes promising guaranteed returns. These are scams.

The bottom line

Use AI like you’d use any online service: share sensibly, check your settings, and double-check anything important. Do that and you can enjoy the benefits with peace of mind.

Our free AI Setup Checklist includes all of these safety steps.