Researchers say AI browsers are a cybersecurity mess

Just a heads up, if you buy something through our links, we may get a small share of the sale. It’s one of the ways we keep the lights on here. Click here for more.

Your web browser just got a brain, and a bit of an attitude. Last week, OpenAI and Microsoft cranked the AI browser race into overdrive with ChatGPT Atlas and a new “Copilot Mode” for Edge. 

These souped-up browsers can answer your questions, summarize web pages, and even do things for you, like a digital intern who never sleeps. 

The idea is a hands-free internet experience where your browser thinks so you don’t have to. The problem? 

It might also overshare, misbehave, or get tricked into doing something very dumb, like giving hackers your credit card number.

The new wave of “agentic browsers” isn’t just about convenience. It’s a full-blown turf war over who controls the front door to the web. 

OpenAI and Microsoft want Atlas and Edge to become your digital butlers, but they’re not alone. 

Google’s baking its Gemini AI into Chrome, Opera’s got Neon, The Browser Company is experimenting with Dia, and upstarts like Perplexity and Sweden’s Strawberry are hustling for a slice of the action.

But as the features pile up, so do the bugs. 

Researchers have already found flaws in Atlas that could let attackers hijack its memory, which is the part where it remembers what you do online, and inject malicious code. (Via: The Verge)

Comet has its own issues, including a vulnerability where hidden instructions could secretly reprogram the browser’s AI. 

Everyone from OpenAI’s security chief to indie researchers admits this is a frontier problem, code for “we have no idea how to fix it yet.”

Experts say these browsers are walking privacy nightmares. They know what you search, read, buy, and say, and they remember it all. 

That makes them hacker goldmines. Worse, the AI agents themselves can be manipulated through sneaky “prompt injections” hidden in text, images, or websites. 

Once compromised, they can click bad links, hand over private data, or change your shipping address mid-purchase.

The verdict? AI browsers are exciting, chaotic, and maybe a little too eager for their own good. 

Until they grow up, experts suggest you keep them on a short leash, or, better yet, turn the AI off and do the thinking yourself.