AI chat tools such as ChatGPT, Gemini and Copilot can draft an email, explain a long letter or help you practise a language, including in Arabic (see ChatGPT in Arabic). Using them safely needs no technical knowledge, only a few habits: be careful what you type, set the privacy options once, check important answers, and watch for fakes.

The settings and rules below come from each company's own help pages and from official Gulf sources, as we checked them on 4 November 2026. Menu names change often.

The short answer

  • Treat what you type as something others might read. Keep ID numbers, health records, bank details, passwords and work secrets out of public AI tools.
  • Change two or three settings once. Each major tool lets you stop your chats being used to train its models, and offers a temporary or history-off mode.
  • Check anything that matters. AI tools can state wrong facts with full confidence. Use them for a first draft, not a final answer.
  • Download apps only from official pages, and check any urgent request for money another way.
  • Children have age limits. ChatGPT is not meant for under-13s.
  • At work, follow your employer's rules and use the tools it has approved.

1. What not to share

Google's privacy page for Gemini says that "a subset of chats are reviewed by human reviewers", that reviewed chats can be kept for up to three years, and asks users: "Please don't enter confidential information that you wouldn't want a reviewer to see" (Google Gemini help).

The UAE government's AI Office made the same point in its May 2026 guide to generative AI applications: "every prompt, document, or dataset entered into these systems poses a potential risk of information exposure or data leaks", so "it is essential to always protect personal and sensitive data" (UAE AI Office, Leading Generative AI Applications; Gulf Today, 10 May 2026).

Keep these out of public AI tools:

  • Identity numbers: Emirates ID, Iqama, national ID, passport and visa numbers.
  • Health information: test results, diagnoses and prescriptions, yours or a relative's.
  • Money: card numbers, bank account details, one-time passwords (OTPs) and login details.
  • Work secrets: client lists, contracts and internal documents, unless your employer has approved the tool.
  • Other people's details: a friend's photo, phone number or private messages.

What to do: if you need help with a document that contains personal details, remove or replace the names and numbers before you paste it in.

2. Change your privacy settings once

ChatGPT (OpenAI). On personal Free, Plus and Pro accounts, sharing your chats to improve OpenAI's models is switched on by default. You can turn off "Improve the model for everyone" and still keep your chat history (OpenAI, Data controls in ChatGPT). The setting is under Settings, then Data controls. After you opt out, OpenAI says it won't use your new conversations to improve its models (OpenAI, How your data is used). For a one-off private question, use Temporary chat: it stays out of your history and is not used to improve OpenAI's models while it remains temporary, though OpenAI "may keep a copy of a temporary chat for up to 30 days for safety purposes" (OpenAI, Temporary chat).

Gemini (Google). The setting is called Keep Activity. When it is off, your future chats won't appear in your activity and won't be used to train Google's AI models unless you send feedback. Google says chats held with Keep Activity off, and temporary chats, "are retained with your account for 72 hours" (Google Gemini help).

Copilot (Microsoft). Microsoft's privacy FAQ says you "can opt out of use of your conversation activity for model training at any time", that conversation activity is stored for 18 months by default, and that you can delete single conversations or your whole history at any time (Microsoft, Privacy FAQ for Microsoft Copilot). The training switches are called "Training on conversation activity" and "Training on voice conversations", under Privacy in your profile (Microsoft, Copilot privacy controls). Microsoft notes that both pages describe the older version of the Copilot app; a new version has been available since 18 August 2026.

What to do: in each tool you use, turn off model training if you prefer, and use the temporary or history-off mode for anything personal. This does not make a chat secret, so section 1 still applies.

3. Check the answers

AI chat tools sometimes produce answers that sound right but are wrong. The US National Institute of Standards and Technology (NIST) calls this "confabulation": "the production of confidently stated but erroneous or false content (known colloquially as 'hallucinations' or 'fabrications') by which users may be misled or deceived" (NIST AI 600-1, July 2024). See hallucination in our glossary.

OpenAI says the same of its own product. ChatGPT can give incorrect dates or facts, make up quotes, studies or references to sources that do not exist, and "may express high confidence even in incorrect answers". Its advice is to "use ChatGPT as a first draft, not a final source" and to verify quotes, data and references (OpenAI, Does ChatGPT tell the truth?).

What to do:

  • For anything about health, money, law, visas or official procedures, check the answer with the official source (a ministry, your bank, your doctor) before you act.
  • Ask the tool for its sources, then open the links yourself. A reference that does not open, or does not say what the tool claimed, is a warning sign.

4. Fake apps, scams and deepfakes

Fake apps. Criminals use the popularity of AI tools as bait. In October 2024 Kaspersky reported that attackers were "using a fake ChatGPT application as bait" to plant a backdoor in organisations in Saudi Arabia. Its advice: "Always try to download software from the official website of the company or service that you are using" (Kaspersky, 9 October 2024). Check the publisher's name before you install.

Deepfakes. A deepfake is a fake video, picture or voice made with AI to look or sound like a real person. Saudi Arabia's data and AI authority, SDAIA, gives consumers three checks in its Deepfakes Guidelines (version 2, May 2025) (SDAIA, Deepfakes Guidelines):

  • Source: check that the message really comes from the organisation or person, using their official website or accounts, or contact them through a channel you already know.
  • Context: ask whether it fits how that source normally behaves. SDAIA notes that public officials "usually do not endorse investment schemes or make unusual offers".
  • Aim: be cautious of messages that create urgency or strong emotion and push you to click a link, download an app or share personal information. If in doubt, talk it over with friends or family.

Our step-by-step guide, How do I spot an AI scam and report it?, covers what Gulf scams look like now and what to do in the moment. If you make pictures with AI yourself, see How to make images with AI.

What to do: never share an OTP, password or card number because of a call or message, however familiar the voice. Call back on a number you find yourself.

5. Children and teenagers

The companies set minimum ages in their own rules:

  • ChatGPT: OpenAI says ChatGPT "is not meant for children under 13", and that children aged 13 to 18 need a parent's consent. It also advises "caution with exposure to kids, even those who meet our age requirements" (OpenAI, Is ChatGPT safe for all ages?).
  • Gemini: Google says you must be 13, or the applicable age in your country, to use Gemini with a personal or school Google Account, and 18 or over with a work account (Google Gemini help).

What to do: set up a child's AI use together and agree the same rules as above: no personal details, and check what it says. Our guide Is AI safe for my child? goes further.

6. Using AI at work

Your employer's policy comes first. If your organisation has approved particular AI tools, use those for work.

The UAE AI Office's May 2026 guide, written for organisations and individuals alike, sets a careful default: "Any data uploaded should be classified as open data unless the platform has undergone a comprehensive assessment by the entity's cybersecurity or data governance team" (UAE AI Office). In plain terms: unless your organisation has checked and approved a tool, only put in it what you would be happy to publish.

What to do: ask your manager or IT team which tools are approved and what data is allowed in them. Read our guide How do I start using AI at work, safely? for a step-by-step start.

7. Your rights under Gulf data laws

Most Gulf states have a general personal-data law: the UAE's Federal Decree-Law No. 45 of 2021 (UAE government portal), Saudi Arabia's Personal Data Protection Law, in force since 14 September 2023 (SDAIA guide), Qatar's Law No. 13 of 2016 (Qatar Tribune; DLA Piper), Bahrain's Law No. 30 of 2018 (Bahrain Legislation and Legal Opinion Commission) and Oman's law issued by Royal Decree 6/2022 (Qanoon). In Kuwait, the CITRA Data Privacy Protection Regulation, as amended by Decision No. 26 of 2024, applies only to licensed telecom and internet providers (DLA Piper; Chambers). These laws apply when an AI tool handles your personal data, as they do to any other service. What they let you ask for, and where to complain, is set out in Your data rights when AI is involved.

8. Where to report a problem

Act quickly. Call your bank first if money or card details are involved, and keep screenshots, numbers and links.

  • UAE: the government portal lists the Ministry of Interior's eCrimes platform in the MOI UAE app, Dubai Police's eCrime website, Abu Dhabi Police's Aman service, the federal Public Prosecution's My Safe Society app, your nearest police station, or 999 (UAE government portal).
  • Saudi Arabia: SDAIA's Deepfakes Guidelines tell victims to report to the platform first, then to submit a complaint through the Kollona Amn app or to the Ministry of Interior's Cybercrime Unit, online or at a police station. For financial fraud or identity theft, they advise reporting to your bank and through the Saudi Central Bank's e-service (SDAIA, Deepfakes Guidelines).

The channels for Qatar, Bahrain, Kuwait and Oman are in How do I spot an AI scam and report it?.

What we don't know yet

  • How the new Copilot app lays out its privacy settings. Microsoft's detailed help pages that we could read describe the version of the app before 18 August 2026. We will update this page when we have checked the new one. If you still use the older app, the same FAQ asks you to sign in to the updated app by 31 December 2026 to keep your eligible chats; content that has not moved over will then be deleted (Microsoft, Privacy FAQ for Microsoft Copilot).

A short checklist

  1. Leave out ID numbers, health details, bank details, passwords and work secrets.
  2. Turn off model training in each tool you use, if you prefer, and use temporary chat for personal questions.
  3. Check important answers with an official source, and open the links the tool gives you.
  4. Install AI apps only from the company's own site or official store listing.
  5. Treat urgent requests for money or codes as suspect, even in a familiar voice. Call back on a number you know.
  6. Follow the age rules, and set up children's use together.
  7. At work, use only approved tools and follow your employer's policy.
  8. If something goes wrong, call your bank, keep the evidence and report it.

Further reading

We checked each setting and source on the company's or authority's own page on 4 November 2026, unless stated otherwise.