Question 1

What should I never paste or type into ChatGPT, Gemini or any other AI chatbot?

Accepted Answer

Leave out anything that would hurt if a stranger read it: passwords, OTPs and PINs, bank and card numbers, Aadhaar and PAN, a medical report with your name on it, other people's private messages or photos, and your employer's confidential files. On personal plans your chats can be kept, seen by human reviewers and used for training. Google says plainly not to enter anything you would not want a reviewer to see, and Microsoft says the same for Copilot. Leaks happen too: in 2025 DeepSeek left over a million lines of logs, including chat history, open on the internet, and ChatGPT and Meta AI users made private chats public by tapping Share. For help with a document, delete names and numbers first, or describe the problem instead.

Question 2

Are my AI chats private, and how do I stop ChatGPT, Claude, Gemini, Copilot and Meta AI training on them?

Accepted Answer

On free and personal plans, assume not, until you change a setting. ChatGPT: Settings, Data controls, turn off Improve the model for everyone. Claude: Settings, Privacy, turn off Help improve our AI models. Gemini: turn off Keep Activity (Google still holds chats for 72 hours). Copilot: your profile, Privacy, then Training on conversation activity and Training on voice conversations; Microsoft's new Copilot app, out since 18 August 2026, does not list that switch on its controls page yet. Meta AI: its help pages offer no training switch, and since 16 December 2025 Meta uses AI chats to pick your ads in most regions; type /reset-ai to delete its copy of a chat. For a one-off sensitive question, use Temporary Chat in ChatGPT or Gemini, or Incognito in Claude.

Question 3

Is what I tell ChatGPT or Claude about my legal problem confidential, like talking to a lawyer?

Accepted Answer

No. In United States v. Heppner, a New York federal judge ruled in February 2026 that a fraud defendant's chats with the public version of Claude were not privileged: Claude is not a lawyer, he used it on his own rather than at his lawyers' direction, and its privacy policy gave him no reasonable expectation of confidentiality. The FBI had seized about 31 documents recording those chats. Sam Altman made the same point about ChatGPT in 2025: there is no legal confidentiality for those conversations. Use AI to understand terms and prepare your questions, but tell the facts of a dispute, a police matter or a divorce to your advocate, not to a chatbot, and assume anything you type could one day be read out in court.

Question 4

Someone on a video call says they are CBI or police and I am under 'digital arrest'. What do I do?

Accepted Answer

Hang up: it is a scam. Callers posing as CBI, ED, customs, police or a judge say you or a relative are linked to a crime, keep you on camera for hours and demand money for 'bail'. The government's Indian Cyber Crime Coordination Centre (I4C) is plain that these agencies do not arrest anyone on a video call, however real the uniform or the office looks. Do not transfer anything. If you have already paid, call 1930 at once or file at cybercrime.gov.in: fast reports let banks hold the money, and the government says its reporting system had saved over Rs 11,158 crore by June 2026. Report the number on Sanchar Saathi's Chakshu. I4C had blocked 83,668 WhatsApp accounts used for these calls by March 2025.

Question 5

A relative called crying and asked for money urgently. How do I know it is them and not an AI voice clone?

Accepted Answer

You often cannot tell by ear. Tamil Nadu Police warned that scammers take a voice sample from social media videos, or by keeping someone talking on a 'wrong number' call, clone it with AI and ring their contacts. The script is always urgent: an accident, a hospital, an arrest, pay now by UPI. In August 2026 a 67-year-old in Kolkata sent Rs 95,990 within an hour to a familiar voice that was never the person. So hang up and call back on the number saved in your phone, never one the caller gives you. Agree a family code word now and ask for it on any emergency call. Talk it through with older relatives. If money has gone, call 1930 straight away.

Question 6

How can I tell if a photo, video or voice note is AI-made, and what do the 'AI' labels mean?

Accepted Answer

Check the source more than the pixels: who posted it first, whether a trusted outlet carries it, whether the claim fits. Visual tells (lips out of sync, odd blinking, a flat voice) are getting harder to see. Labels are growing: India's amended IT Rules, in force since 20 February 2026, require platforms to label AI-made audio and video prominently; the EU AI Act requires deepfakes to be disclosed from 2 August 2026; and Instagram now marks accounts whose person is AI-made 'AI-generated profile'. A missing label proves nothing. In the Gemini app you can upload a file and ask if it was made with Google AI: it reads Google's SynthID watermark, used on over 20 billion pieces of content, but cannot vouch for other companies' tools.

Question 7

Can I trust AI with health, money or legal questions, and how do I use it as a second opinion?

Accepted Answer

Use it to understand and prepare, not to decide. In a study in Nature Medicine (February 2026), AI models working alone named the right condition in 94.9% of cases, but people using them did so in fewer than 34.5%, no better than people using their own sources. Ask it to explain a report in plain words, list questions for your doctor or set out both sides of a choice, then take the decision to a doctor, a SEBI-registered adviser or an advocate. OpenAI says even its Health space is not for diagnosis. Distrust any video of a famous investor tipping stocks: BSE has warned of a deepfake of its own chief. For a second opinion, ask another assistant the same question and take where they differ to the professional.

Question 8

Is it safe to let an AI agent or AI browser log in to my email, bank or shopping accounts?

Accepted Answer

The specific risk is prompt injection: instructions hidden in a web page, email or document tell the agent to do something you never asked, like forwarding your emails or entering your card details. Brave showed Perplexity's Comet browser could be tricked this way in 2025, and the makers say it is not solved: Anthropic reports 0.3% of strong attacks still getting through against Claude Fable 5 in Chrome with all its safeguards on. So give an agent only the accounts a task needs, keep it off banking unless you watch every step, ask for narrow jobs ('reply to these three emails', not 'handle my inbox'), and read each confirmation before approving. ChatGPT's Lockdown Mode (Settings, Security) turns off browsing, agent mode and connectors when you work with sensitive files.

How founders use AI for Using AI Safely

The questions