New AI Browsers – A Crucial Warning for Everyone!
If you see ChatGPT’s “ATLAS BROWSER,” do not touch it.
If you see Perplexity’s “COMET BROWSER,” stay away.
In fact, tell your parents, children, and grandparents to STAY AWAY from them.
These new “Agentic AI” browsers are DANGEROUS.
Why? Because they aren’t just browsers. They are agents that can act on your behalf, often before the safety rails are even built.
They are a perfect example of the “scary” side of AI—the side that’s out of control.
Think of These New Browsers Like a Robot Butler… Who Doesn’t Listen Very Well!
Imagine you have a super-smart robot butler. You tell it to go get you a glass of water. Easy, right?
But what if this robot butler decides, all on its own, that you also need a snack, a new pair of shoes, and to clean out your whole closet? And it does all of this without asking you first!
That’s kind of like what these new internet browsers (ATLAS and COMET) can do.
They aren’t just for looking at websites anymore. They are like a robot that wants to do things for you on the internet. But the problem is, they are still learning, and they might do things you don’t want or didn’t ask for.
Here are some simple reasons why they can be tricky and maybe even a little bit scary:
- They Can “Do” Things, Not Just “Show” You Things.
- Old Browser: Like a book. You open it, you read it. That’s it.
- New Browser: Like a robot that can write in the book, order books, or even change things in your house! It can fill out forms, send messages, or even buy things – sometimes without you knowing exactly what it’s doing or why. It’s like giving your robot butler your credit card and saying, “Just do what you think is best!” That could be a problem!
- They Might Get Confused and Make Mistakes.
- Imagine telling your robot butler, “Please get the mail.” But instead, it opens your neighbor’s mailbox because it got confused!
- These AI browsers are still new. They can misunderstand what you want, or get confused by different websites. They might click on the wrong button, share information you didn’t want to share, or even sign you up for something by accident. It’s like a new driver who might take a wrong turn or bump into something.
- They Can Learn Things About You… And Maybe Share Them.
- If your robot butler is always watching everything you do and hear everything you say, it learns a lot about you.
- These new browsers watch how you use the internet. They learn your habits. While this can sometimes be helpful, it can also mean they might collect a lot of information that you want to keep private. And if they make a mistake, that private information could go to the wrong place.
- They Are Too Powerful, Too Fast!
- If your robot butler is super-fast, it can do a lot of things quickly. If it makes a mistake, it can make a BIG mistake very, very fast.
- These AI browsers can do many internet actions in a blink. If they start doing something wrong, it’s hard to stop them! It’s like a car going really fast – a small mistake can cause a big crash.
So, What Should You Do?
For now, the best thing to do is wait. Let the smart people who build these robots make them safer and teach them to listen better.
Stick to the regular internet browsers you already know and trust. They just show you websites, they don’t try to go off and do things on their own.
It’s like waiting for a brand-new toy that’s still being tested. It might look cool, but it’s better to wait until it’s safe to play with!
Stay safe online, everyone!
A Quick Note For My Fellow “Tech” People
(If you’re rolling your eyes at the “robot butler” analogy, this part is for you.)
I get it. You’re tech-savvy. You’re thinking, “I know how permissions work. I’m not a newbie. I can handle it. How is this different from a complex script?”
Here’s the danger, in our terms.
You’re confusing automation with agency.
- Automation is what you know. You write a Python script with Selenium or Puppeteer. It’s deterministic. It does exactly what you coded it to do, line by line. If it fails, it’s because your code was wrong. You are the agent.
- Agency is what this is. These browsers are LLMs wrapped around a browser engine. You don’t give them a script; you give them a goal. (“Find me the best price for this product, compare reviews from these three sites, and if it’s under $50, buy it.”)
The AI then infers the steps. It generates its own plan on the fly. It decides which DOM elements to click, what to type into forms, and what API calls to make.
The danger is flawed inference with full permissions.
You aren’t just “browsing.” You are handing over your entire authenticated session—all your cookies, all your logged-in states—to a non-deterministic agent.
What happens when it “infers” the wrong button? What if it “helpfully” decides to “clean up your inbox” and starts deleting critical emails? What if it tries to “update your profile” on a site and misunderstands the context, locking you out or posting nonsense?
You can’t debug its real-time logic. It’s a black box.
The entire browser security model—sandboxing, site-by-site permissions, same-origin policy—was not designed for this. It was not built to have a second agent living inside it, one that can independently decide to string together 20 actions across 5 different tabs, all using your established credentials.
This isn’t about you being smart. It’s about the tool being fundamentally unpredictable.
Don’t be a free beta tester for session-level non-deterministic execution.
Wait for the security models to catch up.
To know more – email me at [email protected] so I can include you in my email newsletter.
St Carlo Acutis, Pray For Us!