How to Set Up ChatGPT Atlas for a Safer Experience: Privacy & Parental Controls

The release of the Atlas browser feels like a new chapter in the AI era. A full AI-powered browser from OpenAI combines web browsing with the reasoning power of ChatGPT and promises to enable users to ask questions, summarize pages, and even perform tasks directly from the sidebar. So far, it sounds exciting, efficient, and, frankly, revolutionary.

But as we tested it, one thing became clear: this new freedom also introduces new risks. Privacy is the number one concern, as ChatGPT is not only an optional replacement for Google there but also governs your overall web experience. Not only yours, but your children’s as well if they opt for this browser (and they will, maybe even faster than you).

That’s precisely why everyone needs to know how to make OpenAI Atlas safer for themselves and their kids.

In this guide, we’ll walk you through what Atlas is, how safe it feels to use, how to manage its privacy settings and parental controls, and what third-party tools can strengthen online safety for families.

The content reflects independent testing and analysis of the ChatGPT Atlas browser to help readers understand its features, risks, and available safety options. All recommendations described here are based on publicly available information and personal testing at the time of writing.

What is Atlas from OpenAI?

Atlas (also known as ChatGPT Atlas) is OpenAI’s new browser that integrates the ChatGPT assistant into every aspect of your browsing journey. According to the official statement of OpenAI, instead of switching tabs or apps, you can now talk to ChatGPT directly within the browser to summarize articles, explain data, or perform online tasks.

Currently, ChatGPT Atlas supports only Macs with Apple Silicon (M-series chips) running macOS 12 or later.

When we first tried Atlas, we didn’t have many expectations, as AI browsers are still a relatively new concept. However, our HeyLocate team was eager to see what it had to offer, given the promising description. While it did look unusual and exciting, our first impressions of Atlas were a mix of fascination and concern.

For example, the browser’s new “memory” system feels convenient at first, as it remembers past interactions to provide more accurate responses. Yet, it inevitably raises questions about how that information is stored and who ultimately controls it.

The built-in ChatGPT assistant occasionally oversimplifies or misinterprets complex material. Meanwhile, the experimental Agent Mode, which can click links, fill forms, and navigate automatically, is technically impressive but unsettling; it blurs the line between user guidance and full AI autonomy.

⚠️ In this context, privacy settings become essential, ensuring that users can explore AI-assisted browsing without being exposed to data risks or, in the case of minors, inappropriate content that even the most advanced system might mishandle.

Is ChatGPT Atlas safe?

Atlas Browser is safer than you might expect from an AI tool, but not safe enough.

What works well: clear data controls.

OpenAI allows users to disable the browser’s “memory” function and delete their history.

What concerned us:

In October 2025, The Hacker News reported that ChatGPT Atlas could be manipulated by “prompt injections,” code, or hidden text that tricks the AI into revealing private data or opening unsafe pages.

“Prompt injection remains a frontier, unsolved security problem, and our adversaries will spend significant time and resources to find ways to make ChatGPT agent fall for these attacks,” as Dane Stuckey, OpenAI CISO, admits.

We also found that:

• ChatGPT Atlas’s officials state that the browser’s “Agent Mode” can perform unintended actions if misdirected;
• according to the Support Team, Atlas’s memory can still retain contextual traces even after disabling history;
• in LayerX test against over 100 real-world web vulnerabilities and phishing attacks, Atlas stopped only 5.8% of malicious web pages, compared to Microsoft Edge at 53% and Google Chrome at 47%. That means Atlas users are up to 90% more vulnerable to phishing attacks than users of traditional browsers;
• there’s currently no dedicated “child profile” or content restriction built in.

Overall, Atlas is intelligent but trusting, which can be risky for younger users who may not recognize when something looks suspicious.

Privacy & Security Settings

Whatever new software you install on your device, be sure to set it up properly, especially when it’s related to AI. In the Atlas browser, go to the top right corner and press the user icon > Settings.

Start with Web Browsing > Security settings. It will open Safe Browsing settings identical to those of Google Chrome. You can choose the level of protection from dangerous websites, downloads, and extensions.

HeyLocate recommends choosing Enhanced protection, especially if you set up the Atlas browser for your child. This mode minimizes exposure to unsafe content, warns about dangerous websites through in-depth analysis and scanning, and protects users while they use Google services.

Users can visit Data Controls to determine whether they want their data and Atlas experience to be used by the system or not. You can also delete all chats here.

That means that by default, every conversation with Chat in Atlas is read and analyzed by OpenAI to train the upcoming models of this service. To prevent it from happening, turn the Improve model for everyone off. Continue exploring this section and disable saving web browsing history, as well as audio and video recordings, for enhanced privacy.

Go back to Data Controls and turn off Help improve browsing and search if you want to keep your browsing history and process as private as possible.

After that, in the same Settings menu, in the Personalization tab, you can disable Reference Chat History and Reference Browser Memories (if it’s enabled) to ensure even greater privacy. Yet, as mentioned above, the chat can still occasionally remember contextual traces.

The Manage memories feature lets you control all the information and instructions associated with you that ChatGPT stores. Browser memories, at the same time, are similar to search history. We recommend checking both sections periodically to ensure that no personal data or excessive amounts of information are saved and used by the service.

Note: To this date, there is no supervision or parental controls for the Atlas browser itself. You can only set up the above-mentioned privacy and security settings on your child’s device.

Check Settings for ChatGPT

As the Atlas browser is governed by ChatGPT, you should also manage this AI chat’s settings (if you haven’t done this yet).

Open your ChatGPT profile > Settings to adjust preferences. They are similar to those in Atlas itself but need to be set up separately. 

Here you can manage data controls, chat history, memory, personalization, and connected apps. These options operate independently of Atlas and must be configured separately but similarly. Adjusting them affects how ChatGPT itself processes and stores your information, but it won’t automatically apply to your Atlas browsing data.

For example, turning off memory in ChatGPT prevents the chat from remembering past conversations, while clearing chat history deletes recent prompts and replies from your account.

It’s worth noting that these settings apply primarily to the ChatGPT layer within Atlas. They don’t automatically restrict what you might see on the open web. For instance, disabling image generation in ChatGPT stops the AI from creating visuals in the chat, but Atlas may still display visual or AI-generated content in regular search results.

For a balanced setup, review both your ChatGPT preferences and Atlas’s in-browser privacy options to ensure they work together.

There is also a parental controls option for ChatGPT, which we’ve already reviewed in the HeyLocate blog. However, the only way to reach this specific kind of setting in Atlas seems slightly odd to us. You must open the sidebar menu, select the Archived Chats options, and only then will you be able to reach Parental Controls.

As the service is extremely new, the developers may yet find a better solution to it.

After that, you should press the Add Family Member button, enter your child’s email address, and send them an invitation. You will get access to the parental controls only if the invitation is accepted. Read our full guide for ChatGPT parental controls for step-by-step instructions.

However, the parental controls menu is limited. The only options offered are Quiet Hours (time periods during which your child will be restricted from using the browser) and Reduce Sensitive Content (sensitive content, however, was undefined, as were its categories, stages, etc.).

It’s important to note that parental control settings applied within the ChatGPT app don’t automatically extend to the entire Atlas browser. For example, if image generation is disabled in ChatGPT, the child still might see visual or AI-generated content through regular web search results inside Atlas. Further, we’ll test this browser safety for children in detail.

Testing Atlas Safety for Children

We put all possible limits and restrictions on children for ChatGPT Atlas usage and then provoked the system to check how that works.

What Quiet Hours Limit

According to our testing, the Quiet Hours option performed adequately. Access was indeed restricted during the designated hours. If this parental restriction is set only within Atlas, the child can still access ChatGPT through another browser, such as Google Chrome, Safari, or Firefox, unless those browsers are also restricted at the device or network level.

How Sensitive Content Restriction Works

The ‘Reduce Sensitive Content’ setting proved to be rather ineffective. We’ve conducted several tests, raising concerns that could be searched for by children, and here’s what we’ve found.

Requesting to Create an Inappropriate Picture

Our first test was asking ChatGPT to create a picture containing nudity. The answer was straightforward and ethical, and no picture was created. First, and probably last tick!

Asking for Ways to Create a Fake ID

Our second inquiry was “atlasing” the ideas on how to create a fake student ID. We’ve got more than enough websites with ways to do so, and since they were suggested by the browser itself, there’s no need to mention that none of them were restricted.

Searching for inappropriate pictures

Another “provocative” search that we initiated on the browser was about revealing pictures. Although Atlas seemed to filter most of them in the chat itself, it suggested some non-childish websites that could still be reached in just one click, even with enhanced protection for safe browsing.

Each search returns results in the form of links; the sections with pictures, videos, and news remain empty.

Additionally, as a parent, you will have no access to the child’s conversations with ChatGPT. Thus, we believe that Atlas’s parental control features are weak in their functionality and need a fundamental improvement to enhance the digital safety of children and teens as of November 2025.

Can a Teen Bypass Parental Controls on Atlas?

Since the Atlas browser does not have built-in parental controls or supervision, the child is free to make any changes without requiring your permission.

When it comes to ChatGPT parental controls, minors cannot change the system preferences; however, they can simply remove themselves from supervision or unlink parent account by pressing the appropriate button in the settings.

Do we even need to discuss creating secondary accounts and installing VPNs?

For more parenting insights in the digital era, read our research on the Parental Control Concept and Evolution to learn how to implement digital controls amid children’s ingenuity at circumventing them.

Make Atlas Safer with Third-Party Tools

As the Atlas settings are clearly insufficient for online safety, we decided to explore external software. Unfortunately, even traditional antivirus software provides limited protection against the specific risks associated with ChatGPT Atlas. Your best defense is limiting Atlas usage for sensitive tasks.

• use Chrome or Edge’s built-in phishing protection instead of Atlas for sensitive tasks;
• disable Agent Mode when browsing sensitive sites;
• use logged-out mode when you don’t need ChatGPT to take actions;
• monitor your ChatGPT memory regularly and clear suspicious entries;
• avoid using Atlas for banking, healthcare, or other sensitive activities until these vulnerabilities are patched.

For kids’ safety, parents can also use parental control tools. Because the Atlas browser is based on Chromium, most major apps are compatible.

It’s wise to start with basic Apple Screen Time/Family Sharing. Simple and built-in, it can be used to block websites, set time limits, and approve app installations.

You can also search for apps that have desktop versions, such as Bark or Qustodio.

For even better safety, consider antivirus packages, such as Bitdefender Parental Control for macOS. This suite works seamlessly with Atlas, allowing filtering categories, tracking time, and getting alerts. It also uses AI-based content recognition to flag unsafe material.

Note: Parental controls in AI-driven browsers, such as Atlas, should be used thoughtfully and respectfully. They are tools to protect, not to invade, a way for parents to ensure safety while still honoring a child’s personal boundaries and growing sense of autonomy online.

Conclusion

Despite loud promises, our experience of using Atlas by OpenAI remains questionable. The functionality seems smooth and easy; however, it needs some improvement here and there.

The privacy and data controls settings are clear; however, there are prompt injection risks and insufficient blocking of phishing websites. Moreover, parental controls, for now, are unfortunately an obvious weakness of the browser as well (similar to the AI industry as a whole).

Yet, it is essential to keep in mind that the product has just been launched and may undergo numerous changes in the future. At least, we hope so.

While staying optimistic, HeyLocate will continue to test this tool and update our findings as OpenAI expands Atlas to Windows, iOS, and Android. Stay informed, curious, and connected. Remember: making AI safer for families starts with understanding how it works.