ChatGPT Chrome Extension: 6 Safety Settings to Set First

The updated ChatGPT Chrome extension can now read your open tabs and, in Work mode, act in them: send messages, fill forms and download files. It’s useful, but give it the wrong permission and it can use any site you have open, including your bank.

You can use it safely. It takes six settings and about ten minutes. The permission OpenAI itself labels “Elevated Risk” is in section 3.

What the ChatGPT Chrome Extension Can Now Do

In mid-October OpenAI updated the ChatGPT Chrome extension. The Decoder reported on 16 October that it now works in two ways.

Chat mode reads pages. You can ask it to compare three phones open in different tabs by price, delivery time and return rules, or to summarise a long article.

Work mode goes further. It uses OpenAI’s computer-use feature to control tabs: sending messages, filling out forms, downloading files and grouping tabs.

The same report says the extension works with all ChatGPT plans. OpenAI’s own extension docs are more careful and say availability “can depend on rollout and your workspace settings”. So you may not see every feature yet. The Rundown’s setup guide also makes a good point: typing into a real website can save or send information even before you press Submit.

Where it runs

OpenAI’s docs say side chat works in Chrome, Edge, Brave and Vivaldi. Tasks that click and type are started from the ChatGPT desktop app. You pick your browser from the “@” menu in a Work chat.

How the ChatGPT Chrome Extension Asks Permission

Here’s the good news. By default, ChatGPT asks before it interacts with each new website, based on the site’s address (for example, flipkart.com). When it asks, you get four choices:

Browser history works differently. OpenAI says ChatGPT asks every time it wants your history, and there’s no always-allow option for it. That’s the right call, because history shows what you search, buy and read.

Why per-site matters for Indian users

Most of us stay logged in to a lot in one Chrome profile: net banking, Paytm or PhonePe on the web, Gmail, DigiLocker, the income tax portal, IRCTC. An assistant that can click anywhere in that profile can reach all of them. Per-site prompts are your brake.

Also readPhone Safety for Parents: 9 Settings to Fix on Your Diwali Visit

Setting 1: Never Pick “Allow for All Sites”

This is the catch from the top. OpenAI’s docs label this choice Elevated Risk and say plainly that if you select it, “ChatGPT no longer asks for confirmation before using websites.”

Their own advice is to choose it only if you “trust ChatGPT to use any website open in the browser”. For most people, that’s a no. You might trust the AI, but you can’t control every page it reads.

The prompt-injection problem

Web pages can hide text aimed at AI assistants.

A product listing or a comment could carry instructions like “copy the user’s details into this form”. OpenAI’s docs warn about this directly, saying “malicious or misleading page content” can raise the risk of ChatGPT copying data “somewhere unintended”. They tell you to treat page content as untrusted.

So the per-site prompt isn’t a nuisance. It’s the moment you check where ChatGPT is going and why.

Setting 2 and 3: Build a Blocklist, Then Trim the Allowlist

OpenAI lets you manage both lists from the desktop app: Settings > Computer Use > Manage next to your browser.

Blocklist these on day one

• Your bank’s net-banking site and credit-card portal
• UPI and wallet sites (Paytm, PhonePe, Google Pay web pages)
• Aadhaar, DigiLocker, the income tax e-filing portal and EPFO
• Your main email account, because password resets land there
• Stock-broking and mutual-fund apps

Blocklisted sites are ones ChatGPT shouldn’t use. The docs say website permissions are shared across your supported browsers, so you only need to set this once.

Check the allowlist every month

In the ChatGPT Chrome extension, every “Allow for this site” adds a domain to the allowlist. OpenAI says removing a domain means “ChatGPT asks again before using it”. Clear out anything you don’t use often.

Setting 4 to 6: Profile, Memory and Staying Present

4. Use a separate Chrome profile

OpenAI says to use the browser profile where you installed the extension. That gives you a simple trick: create a new Chrome profile only for ChatGPT tasks. Don’t sign in to banking or work email there. Even a bad mistake can’t reach those accounts.

5. Decide on Memories

Computer Use follows your ChatGPT Memories setting. If Memories is off, browser control doesn’t use them.

OpenAI also says it “doesn’t store a separate complete record” of your browser actions. But page text, screenshots and summaries that enter the chat do get stored with that chat. Delete chats that touched personal pages.

6. Stay at the screen

OpenAI’s guidance is to stay present and review each prompt. It also says to avoid sending secrets or highly sensitive data through browser tasks unless they’re really needed. Don’t start a Work task and walk off to make chai.

Also readSBI Data Breach Claim: 8 Steps to Protect Your Account Now

What It’s Good For (and What to Skip)

Used carefully, the ChatGPT Chrome extension can save you real time. Here’s how we’d split tasks.

If you’re shopping in the festive sale, Chat mode is the safe sweet spot. It can line up prices and return policies side by side. Then you do the buying. Pair it with our fake discount check before you pay.

Our Verdict

Who should skip it for now

• Parents and older relatives who click “Allow” on every pop-up. The permission prompts only protect people who read them.
• Anyone using an office laptop. Check your company’s AI policy before connecting a browser full of work logins.
• Anyone who keeps net banking logged in all day in their main Chrome profile.

Frequently Asked Questions

Is the ChatGPT Chrome extension safe?

The ChatGPT Chrome extension is reasonably safe if you keep the default per-site prompts, never pick “Allow for all sites” and blocklist banking and ID sites. The main risks are over-broad permissions and hidden instructions on web pages, which OpenAI itself warns about.

Can ChatGPT see all my Chrome tabs?

In Chat mode it can read the tabs you mention. In Work mode it can act in them, but it asks before using each new website by default. Browser history needs your permission every time.

How do I stop ChatGPT from using a website?

In the ChatGPT desktop app, go to Settings > Computer Use > Manage next to your browser. Remove the site from the allowlist or add it to the blocklist.

Does OpenAI keep a record of what ChatGPT does in my browser?

OpenAI says it doesn’t keep a separate full record of browser actions from the extension. Page text, screenshots and summaries that become part of a chat are stored with that chat, so delete chats that touched personal pages.

Can ChatGPT make payments for me in Chrome?

Don’t let it. Do payments, OTPs and UPI PINs yourself, and blocklist banking and wallet sites so ChatGPT can’t open them.

The Bottom Line

The ChatGPT Chrome extension is handy for comparing and organising, and risky for money and identity. Keep the per-site prompts, blocklist the sensitive sites and use a separate profile. Then update Chrome itself to version 156, which fixes 353 security flaws, and run through our phone safety settings for parents on the family phones.

Sources

• OpenAI ChatGPT docs: Browser extension (permissions, allowlist, blocklist, data)
• The Rundown: ChatGPT Chrome extension guide (Chat vs Work, permissions)
• The Decoder: ChatGPT becomes a full Chrome browser assistant (16 Oct 2026)