AI-powered browsers claim to simplify our online lives. But experts warn that their promised convenience comes at the cost of privacy, transparency, and the open web itself.
Perplexity’s Comet and ChatGPT’s Atlas are just the latest examples of a growing number of AI companies embedding large language models (LLMs) directly into web browsers. This combination promises everything from conversational web page summaries and contextual answers to automating tasks like planning travel and making reservations.
At first glance, these features read like useful convenience upgrades. But beneath the sales pitch, digital-rights and privacy experts see something else taking shape. They warn that AI-native browsers could reshape the internet’s architecture, weaken privacy protections, and hollow out the open web.
“This is Web 4.0, it locks us away from the open web entirely to whatever the large language model hallucinates is the information,” said Esther Payne, a privacy advocate and community manager at the Librecast Project, in a chat with Cybernews.
The new gatekeepers
Browsers have long been treated as neutral conduits to websites. However, the experts we spoke to argue that neutrality is already conditional, and adding an AI layer will turn the browser into an intrusive middleman.
Alexander Linton, a digital rights advocate and president of the Session Technology Foundation, told Cybernews that the web has always had sentinels, from domain registries to certificate authorities, but AI changes the stakes.
“Although the web already has gatekeepers and vulnerabilities to censorship, the adoption of AI browsers would make the situation significantly worse,” he said.
The problem, he explained, isn’t just about access – it’s about interpretation. Once the browser begins deciding what to show or summarize, it becomes a filter for reality.
“When they [AI agents] are handling sensitive information, such as banking or health data, the model may hallucinate and perform actions that you did not request.”
Payne sees the rise of AI browsers as the next stage in the caging of the web, an evolution from Web 2.0’s curated communities and walled gardens, to a Web 4.0 where users browse a version of the web pre-filtered by an AI model deciding what they should see.
Jürgen Geuter, founding member of the Otherwise Network think tank and a known critic of technological fads, offers a simpler metaphor. In an email exchange with Cybernews, he refers to the browser as “the operating system” for our internet-existence,” and argues that by embedding AI at its core, companies aren’t just upgrading a tool, but are rather trying to take over the space where we live online.
Trading in privacy
The same contextual awareness that makes AI browsers useful also makes them deeply intrusive. To work their magic, these AI browsers must constantly observe our actions by reading the pages we visit, capturing our inputs, and even remembering what we’ve done previously. For privacy advocates, that requirement turns the browser from a tool of exploration into an instrument of surveillance.
Payne describes a deeper structural risk and questions the very nature of how LLMs process data.
“In order for AI assistants to function, you have to provide unencrypted data about yourself to put into its training data,” she said, calling it a “privacy nightmare.”
She points to real-world evidence of that exposure.
“Prompt injection attacks have shown that it's possible to extract large amounts of data unmodified from the training models.”
And the law, she argues, hasn’t caught up yet. She believes the current privacy laws do not apply here as the data isn't encrypted.
“So the question to be asked is, are the current AI providers legal under data protection laws? I would say no.”
Linton warns that these new-fangled pieces of software demand a level of access that is simply incompatible with privacy-conscious use, and bluntly asks users to “currently, avoid using any agentic AI browsers or technology if you are concerned about maintaining your privacy.”
The vanishing web
Our experts agree that the rise of AI browsers doesn’t just affect privacy, but also threatens to unravel the web’s very structure. By pulling information into proprietary models and delivering it through conversational answers, these systems turn the open web into raw material for closed platforms.
They call this the new rentier model of the web.
“AI companies are fundamentally built on the principles of rentier capitalism,” said Linton.
“AI models were founded using internet commons (publicly accessible books, websites, code), and are now racing to build the largest moat possible to capture the growing industry.”
In this model, our experts argue, knowledge becomes a closed loop. It is scraped from the public web, reassembled by AI, and resold to users who never realize where it came from.
This is why Geuter believes the danger of these AI web browsers is not only the immediate content filtering, but the longer-term dependency they create where users don’t feel the need to visit the web itself.
He warns that this change could harden into habit. Once people grow used to getting pre-digested answers, the instinct to search, click, and verify begins to wear away, and will over time alienate people from the open web.
And when fewer people click through to original sites, publishers lose traffic and revenue. Payne worries this could, in turn, accelerate the decline of independent publishing.
In the end, what began as a way to simplify browsing may end up rebuilding the web in the image of the AI-enabled web browser companies.
Unlock more exclusive Cybernews content on YouTube.
Your email address will not be published. Required fields are markedmarked