Is ChatGPT Safe? Privacy Settings and What Never to Paste into AI

By Ravindra Bagale, trainer · LinkedIn · Last updated: 19 October 2026

Generative AI course · Chapter 29 · Estimated reading time: 11 minutes

1. Suppose we want a renewal reminder for a client

Suppose we are Meera's insurance agent, Rohan, in Kolhapur. We want AI to draft a polite policy-renewal reminder. We paste:

Client: Meera Kulkarni, meera.k@example.com, +91 9876543210,
PAN ABCDE1234F, Aadhaar 2345 6789 0123. Policy expires 30 Nov.
Write a renewal reminder.

(These are dummy values made up for this page.)

The reminder is fine. But we did not need any of the private details to get it. The AI only needed: "a client, a policy, a date". Everything else was risk with no benefit.

2. What happens to your chats

1. Your words go to the company's servers. The AI does not run on your phone. Your text, files and images are sent to the company.
2. Your chat is saved in your history, unless you use a temporary or incognito chat.
3. It may be used to train future models, on many personal (consumer) plans, unless you turn that off.
4. A small part may be read by people. Some companies say trained reviewers look at a sample of chats to improve quality and safety.
5. Business plans are different. Company plans (ChatGPT Business or Enterprise, Microsoft 365 Copilot with a work account, and similar) usually do not train on your data by default. Check your company's agreement.

So two separate questions matter: Is my chat used for training? (you can usually switch that off) and Is my chat stored? (yes, at least for some time, even with training off).

3. The "never paste" list

Never paste these into any AI chat, with any setting:

• Passwords, OTPs, PINs, CVVs and recovery codes
• Aadhaar, PAN, passport and bank account numbers, yours or anyone else's
• Other people's private details: phone numbers, addresses, health information, salary
• Client or company documents marked confidential, or anything your contract says to keep private
• Source code with keys or passwords inside (API keys, database passwords)
• Photos of ID cards, cheques or bank statements

If AI needs the shape of the data, give it a placeholder: [CLIENT NAME], [PHONE], [AMOUNT]. If it needs a number to calculate, use a dummy number of the same size.

4. Worked example 1: turn off model training in 4 tools

These are the steps the companies document in October 2026. Menu names change; if you cannot find a setting, search the tool's help centre for the name in quotes.

ChatGPT

1. Open your profile (account menu) and select Settings.
2. Select Data controls.
3. Turn off Improve the model for everyone.

When signed in, this applies to your account on all devices. Your chats still appear in your history. If you press thumbs up or thumbs down, OpenAI says the whole conversation with that feedback may still be used for training.

Gemini

1. Go to gemini.google.com and open Settings & help, then Activity.
2. Select Turn off, or Turn off and delete activity.

Google says that with Keep Activity off, your future chats are not used to train its AI models unless you send feedback, but they are still kept with your account for up to 72 hours. Chats that human reviewers already looked at are kept separately for up to 3 years.

Claude

1. Open Settings, then Privacy.
2. Turn off Help improve our AI models (earlier called "Help improve Claude").

Anthropic says new chats are then not used for future training, but conversations flagged by its safety systems may still be used for safety work.

Microsoft Copilot (personal account)

1. Open your profile, then Privacy.
2. Turn off Training on conversation activity and Training on voice conversations.

Microsoft released an updated Copilot app on 18 August 2026, and its help pages describe the settings separately for the older and newer app. If you do not see these switches, open Microsoft's privacy page for Copilot and follow the steps for your version. With a work or school account, Microsoft 365 Copilot follows your company's data protection rules instead.

Expected output (yours will be similar, not the same): in each tool, the switch shows as off. Take a screenshot of each setting page for your own record, because updates sometimes add new switches.

5. Temporary and incognito chats

Use a temporary chat for one-off personal questions. It is still not a safe place for passwords or ID numbers. "Temporary" means it disappears from your history, not that it was never sent.

6. Worked example 2: placeholders, with a small Python helper

The safest data is data you never paste. Before pasting, replace private details with placeholders. You can do this by hand, or with a small script for text you paste often.

# Replace common private details with placeholders before pasting text into an AI tool.
import re
RULES = [
    (r"\b[A-Z]{5}[0-9]{4}[A-Z]\b", "[PAN]"),
    (r"\b[2-9][0-9]{3}\s?[0-9]{4}\s?[0-9]{4}\b", "[AADHAAR]"),
    (r"\b[\w.+-]+@[\w-]+\.[\w.]+\b", "[EMAIL]"),
    (r"(?:\+91[\s-]?)?\b[6-9][0-9]{9}\b", "[PHONE]"),
    (r"\b[0-9]{2}[A-Z]{5}[0-9]{4}[A-Z][0-9A-Z]Z[0-9A-Z]\b", "[GSTIN]"),
]
def redact(t):
    for pat, ph in RULES:
        t = re.sub(pat, ph, t)
    return t
sample = ("Client: Meera Kulkarni, meera.k@example.com, +91 9876543210. PAN ABCDE1234F, "
          "Aadhaar 2345 6789 0123, GSTIN 27ABCDE1234F1Z5. Please draft a renewal reminder.")
print("BEFORE:", sample)
print("AFTER: ", redact(sample))

Real output (we ran this on 19 October 2026 with the dummy values above):

BEFORE: Client: Meera Kulkarni, meera.k@example.com, +91 9876543210. PAN ABCDE1234F, Aadhaar 2345 6789 0123, GSTIN 27ABCDE1234F1Z5. Please draft a renewal reminder.
AFTER:  Client: Meera Kulkarni, [EMAIL], [PHONE]. PAN [PAN], Aadhaar [AADHAAR], GSTIN [GSTIN]. Please draft a renewal reminder.

Look closely: the email, phone, PAN, Aadhaar and GSTIN were replaced, but the name was not. Patterns can catch numbers and emails; they cannot reliably catch names, addresses or medical details. So always read the text once before pasting, and replace names by hand: "Client: [CLIENT NAME]".

Then paste only what the AI needs:

Write a polite renewal reminder from an insurance agent to [CLIENT NAME].
Their health policy expires on 30 November. Ask them to renew before that date
and offer a call to answer questions. Under 80 words. Sign as [AGENT NAME].

Expected output (yours will be similar, not the same): a short, polite reminder with the placeholders in place. You fill in the real name in your own email or WhatsApp, not in the AI chat.

7. Worked example 3: check your company's AI policy

Before using AI for work, find out what your company allows. Many companies have an AI policy, sometimes inside the IT or information security policy.

Questions to answer (ask your manager or IT team if the policy does not say):

1. Which AI tools are approved? (For example, only the company's Copilot or ChatGPT Business account.)
2. Can I use my personal AI account for work? (Often the answer is no.)
3. What data is never allowed? (Client data, source code, financial results before they are public.)
4. Do I need to say when a document was drafted with AI?
5. Who checks AI output before it goes to a client?

A prompt to understand a long policy (paste the policy only if it is not itself confidential, or use the company's approved tool):

Here is my company's AI usage policy. Summarise it in a table with 3 columns:
Allowed, Not allowed, Ask first. Then list any rule I might break by mistake
in daily work. Quote the policy line for each rule.

Expected output (yours will be similar, not the same): a 3-column table and a short list of easy-to-miss rules, each with the policy line quoted. Read the quoted lines in the policy itself; do not trust the summary alone.

If your company has no policy, the safe default is: use AI only with public or made-up information, never with client or company data, until your manager agrees in writing.

8. Beginner mistakes and quick fixes

9. Try at home

Recap

Open your AI tool's settings today and check one switch. Then make the "never paste" list a habit, because no setting can protect data you have already sent.

Previous chapter: AI for YouTube and Instagram Reels: From Script to Captions, Plus a 30-Day Plan

Next chapter: Deepfakes Explained: How They Are Made and How to Check a Suspicious Video, Photo or Voice Note

Ravindra Bagale, trainer: linkedin.com/in/ravindra-bagale. The people, company and all ID numbers on this page are made up for learning. Settings and retention periods were checked on OpenAI's, Google's, Anthropic's and Microsoft's help pages on 19 October 2026 and can change. The Python output was produced on 19 October 2026. AI output changes every time, so your output will be similar, not the same.