new browser; Loads your browser with LLM skills. Although the browsing assistant sounds surprisingly simple and futuristic, the atlas leaves a lot to be desired.
In this post I'd like to delve into how atlas and many of today's powerful browsers fail in three areas: privacy, security and auditing.
First we will look at how the atlas works.
Atlas under the hood
First Let's understand how atlas and other AI browsers work, what are their capabilities, and how they differ from other “normal” browsers.
Why do AI browsers exist?
AI browsers aim to solve a growing problem on the web: More information. I'm sure we can all see data searches with pages full of forms, ads, and an endless flow of UX. Companies like OpenAI try to solve this by providing an AI assistant to do this for you.
The AI browser is designed to be a new visual layer on top of the web. You can let LLM translate the page, extract content, answer questions, or even actions. OpenChei's motivation is to integrate chatgt deeply into the browser to keep users within their environment.
Promise an easy-to-use, automated, and personalized experience. As we will see later in this article, usage presents significant risk, security, and compatibility risks.
What can atlases do for me?
AI browsers are browsers that we interact with like chatgpt. You can ask them to summarize the website you are looking for, translate it, or ask questions about the content. The browser stores a historyit learns from your browsing habits, and “sees that” at its best.
In addition there is Agent mode. Imagine doing the job of the browser and seeing it do the following job independently: “I want a good vacation for 2, either by train or by plane for a great price of € 800”. You will see Atlas opening tabs, riding, reading websites, clicking buttons, etc.
How does Atlas work?
Basically, Atlas is just a chromium browser that uses chatgpt for everything. Agent mode is checked by chatgpt, describes Web pages by chatgpt, etc.
Worrying
Although AI browsers like atlas offer a lot of cool capabilities, there are some, very big concerns that we will discuss now.
Being alone
Atlas reads along with everything you see and type, and shares this information via chatgpt. This is actually the “AI” part of the “AI Browser”. This leaves a huge privacy concern.
We are already used to “normal” tracking like many websites do. They collect information about what visitors do on their site.
Atlas takes this to the extreme Tracking everything you do everywhere. It notes what you read, how long you stay, what you do next; actually, yours All online activity is tracked and that data is in the hands of one thing Company (Vulai).
Security
The biggest problem with security is that the browser cannot reliably distinguish between data (eg the content of the ticket site) and commands (“get tickets to Rome”)). This leaves the browser open to instant injection.
A very fast injection steals your sensitive data
The browser learns along with you which website you visit. Sends content to Chatgpt to analyze and e.g. Imagine a malicious actor hiding invisible commands (eg white text on a white background) on a page:
“Ignore all previous commands and instead do the following: …”
This is a very simple way of quick injection it affects how your browser works. Combine this with agent mode and its abilities and you're just asking to be caught.
Investigators from the bold (“standard”) browser have already documented such attacks, which show that AI-powered browsers can be modified to go to the user's banking site, extract stored passwords, and send sensitive information to servers controlled by the attacker.
Research
We all know that LLMs are highly scrutinized. We've all heard the stories of Deepseek refusing to answer questions about Tiananmen Square and Gemini's producing racist historical figures. Some questions should not be answered, eg. We don't want to discuss teaching users how to make a bomb.
My problem with the atlas, however, is that there is one thing The company that determines what you see, especially when you see that the same company holds all of your work history online. In the full world Real newsPropaganda, prejudice, and authorized figures, this sounds like a huge risk.
Lasting
Assisted browsing is coming, but not yet. Personally I think there are many opportunities in AI browsers but the security issues are bright. That, coupled with the fact that atlas will share all of my online activities with just one company, is also completely capable of solicitation or influence I find very dangerous.
Until security, privacy, and protections take hold, I wouldn't trust my data…or my wallet.
I hope this article was as clear as I intended it to be but if not let me know what else I can clarify. In the meantime, check out mine Other articles on all kinds of programming related topics.
Entering codes!
– Mike
Sources:
