Opening my Gmail inbox
In this first part, I decided to test how Atlas interacts with my Gmail inbox, handles calendar notifications, and automates tasks like marking certain senders as read.
I started by opening ChatGPT Atlas and using its search bar to ask it to open my Gmail inbox. The interface responded with a prompt offering to open my inbox for me.
I started by using the ChatGPT Atlas search bar to type “open my Gmail inbox”. Atlas suggested this query in the predictive search results, making it easy to select.
Atlas then asked for my permission to open the inbox via the integrated browser. I clicked Continue, and Atlas proceeded to open Gmail.
Before fully loading the page, it displayed a caution about Agent mode introducing risks when logged in. Since I needed it to access Gmail, I chose Start logged in.
Please pay attention: 👉 Atlas is great for everyday and productivity use — but sensitive work (e.g., banking, regulated data) should still stay on a traditional browser until Atlas’s enterprise controls fully mature. Check more about chatGPT Atlas privacy at the end of this article.
How to Search For Emails and Sort Them Using The Atlas Browser
Once Gmail loaded, Atlas scanned my inbox. I added a prompt asking Atlas to find all the calendar notifications. It looked for calendar invitation emails under the Calendar Notifications label. The inbox contained a variety of messages, and Atlas quickly identified where the calendar invites were stored.
Atlas loaded the Calendar Notifications label, where it found 67 calendar invitation emails. This search took about 45 seconds to complete.
To clean up my calendar invites, I asked Atlas to delete invitations older than one month (before 22 September 2025). It selected 63 messages and moved them to Trash. Gmail confirmed the deletion, and the process finished in about 28 seconds.
How to Create a Gmail Filter with chatGPT Atlas Browser
Next, I wanted to create a filter that would automatically mark emails from Team Tactiq as read.
In my first attempt, I mistakenly searched for “Team Tactic” (with a c), and Gmail reported that no conversations matched. Atlas pointed out the typo and adjusted the filter criteria itself — which is very cool.
It went through the whole process of creating a filter, filled in all the necessary fields, and reported that it was done.
I searched for “Team Tactiq” manually to check if all messages were marked as read — yes, they were.
I also navigated to Gmail Settings and found that exact filter at the end of all the filters I have.
How to Find All Subscriptions Within The Atlas Browser
I asked Atlas to show all my subscriptions. As announced by the ChatGPT team, it keeps context, so users don’t need to clarify everything.
It tried to look for the “Subscriptions” label in my Gmail account but didn’t find it.
Then it searched my browser history but also failed, so it asked for more specific details.
Alright, I used a longer prompt: “Show me all email lists I'm subscribed to.” This time, its search led to opening the Clean Email app, as it recognized it as a tool I use to manage email subscriptions. It kindly asked me to log in.
After logging in, I asked it to open the Unsubscriber tab. I was curious if it could do that — and it could!
I wanted to experiment further, so I clicked on the Atlassian subscription and Clean Email opened the list of all messages from that subscription. I wanted to see how it keeps the context and follows my actions.
The next prompt was more challenging—ChatGPT needed to create a new automated rule that would send all messages older than two months from the sender I was viewing to the Trash.
The first part went well: it identified that the automated rule related to the Auto Clean feature and opened it.
Next, it found the + button and almost created the new Auto Clean rule:
- It added a name for the rule.
- It added a sender (though the sender was incorrect).
- It chose the “Older than 2 months” filter.
- And it selected the Trash action.
At this point, I stopped it when I saw the sender was wrong. I had to update the sender manually, which indicated that the prompts should be more specific.
Here’s a more concise version that keeps all essential facts and verified source links 👇
How Safe and Private ChatGPT Atlas Is
While I’ve been using ChatGPT Atlas, I couldn’t help but feel a bit uneasy about privacy and safety. After all, I’m logging in with my real email — the same one connected to important accounts like my bank — and this new AI agent can actually perform actions on its own.
That’s both impressive and a little scary, so I decided to look into how safe Atlas really is and what kind of privacy protection OpenAI has built around it.
Read more: Email Security Best Practices
What Users Can Control About Safety in ChatGPT Atlas
Good news here. You can control what Atlas remembers about you, decide if your activity helps train AI models, and even manage how much access the built-in AI agent gets.
- Go to Settings → Data controls in the Atlas browser.
- Under “Include web browsing”, you can toggle whether your browsing content is used to train models (it’s off by default).
- Under “Page visibility”, you can disable ChatGPT’s access to specific web pages so they won’t be used as memories.
- For enterprise users: Admins can enable/disable Agent Mode and set permissions workspace-wide.
Everything is encrypted, and OpenAI follows serious privacy standards like SOC 2 Type II and GDPR. That level of transparency is refreshing — they even publish regular trust how few data requests they get from governments.
Privacy and Security Risks in ChatGPT Atlas
That said, it’s still an AI-powered browser, which means some data is processed through OpenAI’s cloud. So if you’re dealing with super-sensitive stuff — like banking, medical info, or corporate data — I wouldn’t rely on Atlas just yet.
Privacy & Safety Risks Across Four Browsers — ChatGPT Atlas, Google Chrome, Safari and Opera
I got curious about how safe ChatGPT Atlas really is compared to browsers like Chrome, Safari, and Opera — so I put together a quick table to break it down.
| Browser | Major Privacy/Safety Strengths | Key Risks / Weaknesses |
|---|---|---|
| ChatGPT Atlas | • Built with AI-assistant features that give you convenience (task automation, memory of past browsing, etc.). • Offers “Memories” and controls for what the browser remembers and whether browsing data is used for model training. | • Because it includes an AI agent that can act, new vectors of attack exist (e.g., prompt injection, agent misuse). • It may store OAuth tokens or browser session credentials insecurely (un-encrypted tokens in macOS). • Less mature in terms of full enterprise audit/assurance; you’re trusting the cloud + vendor heavily. |
| Google Chrome | • Very mature ecosystem, frequent updates, large security resources, strong sandboxing & phishing protections. | • By default, tracking, sync, and linking across Google services may reduce privacy out-of-the-box. • Large market share makes it big target. • Less “opt-in” user-control compared to more privacy-centric browsers. |
| Safari | • Good integration in Apple ecosystem, decent default privacy features (e.g., Intelligent Tracking Prevention). | • Limited built-in VPN or advanced privacy features unless you pay for Apple’s Private Relay. • Safari still relies on web content and Apple’s cloud ecosystem; not zero-trace. |
| Opera | • Offers built-in features like free VPN, ad-blocker/tracker-blocking built in (better than many out-of-box). • Often uses same Chromium base as Chrome, so familiar. | • VPN is free but may have limits and still requires trust in the service provider. • Some security updates may lag behind Chrome’s upstream base. |
My Take & Recommendation
- If you’re using Atlas, the convenience of its AI-agent features is impressive — but you need to be comfortable with giving it more access (to tasks, memory, credentials). That means you should be extra cautious about what you let it control and monitor your settings actively.
- Chrome, Safari and Opera are more traditional browsers: they have fewer “agentic” features (less automation), which also means fewer “novel” attack surfaces.
- For everyday browsing, all four are acceptable — choose based on your comfort level with privacy vs convenience.
- For high-sensitivity work (banking, healthcare, confidential corporate data) I would lean toward using a browser with minimal third-party / AI-agent integration (e.g., Safari with Private Relay or Opera with VPN) and keep the AI-agent-enabled browser separate if you’re using Atlas.
- Always update your browser, review privacy/sync settings, and use separate profiles or browsers for high-risk vs routine tasks.
Read Also: The 9 Best Free Private Email Providers and How Modern AI Email Spam Filters Work.
Wrap Up
Overall, my experiments showed that ChatGPT Atlas can quickly open Gmail, manage calendar invites and filters, and even integrate with Clean Email for deeper Gmail inbox organization. Most tasks took less than a minute, which makes it a surprisingly powerful tool for handling emails.
If you’re using Atlas, its AI-agent features feel incredibly convenient — but they do come with a tradeoff. You’re giving it more access to your tasks, memory, and credentials, so it’s worth being extra careful about what you allow and checking your settings regularly.
