Who’s Surfing the Internet, You or Your Browser?

A few weeks ago I wrote about the advent of AI-powered browsers ushering in an age where software doesn’t just open web sites, it’s starting to observe us, analyzing what we read and taking the initiative for us. Since then, the ecosystem has evolved further, raising more concerns.

Fast Company’s devastating analysis of ChatGPT Atlas, Perplexity Comet and many others shows that while these programs promise to automate everyday tasks, from writing shopping lists to canceling subscriptions through the mystery of the algorithm, they do so at the risk of the browser no longer being a tool at our disposal and, instead, an agent that uses us.

Fast Company says AI-powered browsers exposes us to a new vulnerability: no less than prompt injections, or commands hidden in apparently innocent pages that can manipulate the agent to leak passwords, move money or extract personal information. Researchers at Brave and Malwarebytes have confirmed this in two studies: a simple invisible command can cause a browser like Atlas to execute actions without the user noticing.

OpenAI, aware of the problem, published a privacy guide that tries to explain how Atlas analyzes all the pages we visit to “remember context” and give us more personalized results. In other words, the browser observes, records, and learns from us permanently. The line between assistant and overseer is becoming increasingly blurred.

This brings us to a central question: who do these browsers really work for? The promise of saving us time is integral to a business model based on extracting our behavior, our searches and habits to feed models and advertising. Search Engine Land recently revealed that OpenAI is already testing a version of ChatGPT with advertising, confirming that this “memory” is not just a way of providing us with help, but has, above all, an obvious commercial value.

In parallel, Anthropic has announced Claude for Chrome, and Vivaldi, in the opposite direction, has published “Vivaldi takes a stand: keep browsing human”, a post in which it explicitly defends a browser without built-in artificial intelligence or contextual tracking. The dichotomy could not be clearer: some people are moving toward a navigator that predicts and acts, while the more cautious among us are sticking with those that protect and accompany.

The debate over AI-powered browsers reflects something bigger: the ongoing battle between control and convenience. Every time a technology promises to do something for us, we should ask ourselves what it gets in return. The history of the internet is full of examples where ease of use became a dependency trap. And as Cory Doctorow explains in “Enshittification”, all platforms follow the same pattern: first they optimize for the user, then for the partners, and finally for themselves.

The smart browser is tempting: saving clicks, automating tasks, everything related to a smoother experience. But what do we give up in return? When the browser knows everything about you, the balance between privacy and productivity is broken. The dream of customization ends, as always, in surrender.

Perhaps, as Vivaldi’s manifesto suggests, we need to recover an old idea: exploring, as opposed to being guided. Maybe true intelligence isn’t about a browser that makes decisions for us, and instead helps us make them more clearly. As things stand, AI-powered browsers leave a lot to be desired.

If tomorrow’s navigators cease to be a window on the world and instead merely reflect what we want to see, then the problem will no longer be technical, but philosophical. And when that point comes, if it hasn’t already arrived, we’ll have to ask ourselves who really surfs the internet: us, or AI.

—

This post was previously published on Enrique Dans’ blog.

***