OpenAI launches the ChatGPT Atlas browser: a privacy nightmare | Press review n°29
In this weekly press review about artificial intelligence: ChatGPT Atlas, Amazon planning to replace 600,000 jobs with robots, and the existential risks of superhuman AI.
📰 Read
OpenAI launches the ChatGPT Atlas browser: a privacy nightmare
The company behind ChatGPT rolled out a new web browser called Atlas. It allows the conversational agent to analyze and remember the websites visited by users. Experts warn that this creates serious risks related to privacy, The Washington Post reports.
Atlas is currently available on macOS only. It is built on the open‑source Chromium browser, which was created by Google in 2008. That same base code underlies browsers such as Google Chrome, Microsoft Edge, and Brave.
What sets Atlas apart from its competitors is the integration of ChatGPT directly into the web browsing. The home page is a chat window where you can ask a question to ChatGPT or enter a url. With this feature, Atlas changes the fundamentals of internet navigation: instead of typing a few words into a search engine like Google or DuckDuckGo, Atlas users will primarily rely on the conversational assistant.
The ChatGPT integration doesn’t stop there, it is even omnipresent. By clicking the “Ask ChatGPT” button at the top right corner, users open the chatbot in a sidebar and can request a summary of the currently viewed page or ask for additional information related to their ongoing search.
In a launch video, OpenAI shows the example of a user looking at an offer for a vacation home rental and asking ChatGPT whether there are good hikes nearby.
Introducing ChatGPT Atlas (OpenAI)
When using Atlas, it is also possible to ask ChatGPT to draft or edit an email directly within the browser, without having to copy‑paste the text between separate windows.
Finally, the OpenAI browser includes an agent that can, for example, book a hotel or make online purchases autonomously. This capability is limited to Plus accounts ($20 per month) and Pro accounts ($200 per month).
However, the agent mode creates cybersecurity vulnerabilities, Fortune warns. A type of cyberattack called “prompt injection” allows hackers to hide instructions for ChatGPT on websites. The chatbot can then open the user’s email inbox and send all the messages to a criminal group, for example.
Privacy risks
Atlas also creates new privacy risks for internet users. Geoffrey A. Fowler, a technology columnist at The Washington Post, warns that “behind the scenes, [Atlas] is working to learn much more about you.”
If you grant permission during setup (which most people likely will), the browser “builds a trove of memories about sites you visit,” Fowler notes. And the browser doesn’t just remember website addresses, it also records “facts and insights” from the sites themselves. These memories, stored in OpenAI’s data centers, are then used to customize ChatGPT’s responses to each user.
“That level of personalization brings privacy risks that are hard to understand, much less control,” warns Fowler. “There are things you might want an AI to remember and bring up again in the future — and things you definitely wouldn’t, such as relationship troubles or that embarrassing medical condition you researched at 2 a.m.”
A contradicted statement
OpenAI claims Atlas does not record entire page contents. The browser is also not supposed to remember government IDs, bank account numbers, addresses, passwords, medical records, financial information, or the content of adult websites.
However, a test conducted by the Electronic Frontier Foundation and cited by The Washington Post showed that Atlas kept memories about a registration for “sexual and reproductive health services via Planned Parenthood Direct,” a telehealth platform. The browser also recorded a doctor’s name. According to Lena Cohen, the technologist who performed the audit, “the extensive data collection in the Atlas browser could be a privacy nightmare for users.”
“Previously, there was a fairly visible barrier between what we fed to ChatGPT and our lives outside of it,” summarizes Fabien Suchanek, professor at the engineers school Télécom Paris and an AI specialist, in an article for Libération. “With Atlas, that barrier falls. Even with the ability to ‘switch‑off’ the feature, there’s still a risk that we share more than we intend.”
Serious consequences
“Online surveillance can have serious consequences,” alerts The Washington Post. “Will governments be able to ask OpenAI to hand over people’s browsing data and memories? What if they’re researching activities that are illegal in certain states, such as abortion?,” Fowler wondered.
OpenAI told him that it would only disclose user data through “valid” legal processes or in an “emergency situation.” But if there are backdoors in ChatGPT, OpenAI doesn’t need to disclose information in order for it to be obtained by other institutions anyways.
On this topic: Chatbots are surveillance machines | Press review n°6
The Atlas browser thus represent a further decline of privacy on the internet. A sacrifice that some people will do willingly in order to use new features while they surf the web. In short, we normalize and trivialize the fact that our online activities are increasingly surveilled and recorded.
Let’s keep in mind that this invasive surveillance capitalism is often conducted by companies based in the United States, a country which is moving at great speed toward an authoritarian political regime and where digital surveillance is constantly reinforced, especially on social medias.
ChatGPT’s Altas Browser is a Security Nightmare (Low Level)
Seven important news this week
ICE is building a social media panopticon (The Verge)
A $60 Mod to Meta’s Ray-Bans Disables Its Privacy-Protecting Recording Light (404 Media)
Meta Layoffs Included Employees Who Monitored Risks to User Privacy (The New York Times)
Nobel winners and celebrities challenge Silicon Valley’s vision for the future (The Washington Post)
Anthropic Has a Plan to Keep Its AI From Building a Nuclear Weapon. Will It Work? (Wired)
Sam Altman’s next startup eyes using sound waves to read your brain (The Verge)
🎥 Watch
Amazon plans to replace 600,000 jobs with robots
In a video posted Thursday, political commentator Kyle Kulinski talks about an article in The New York Times revealing that Amazon intends to replace more than 600,000 jobs with robots by 2033 in order to cut labor costs.
It is a worrying development for workers worldwide, who will find it increasingly difficult to find jobs as artificial intelligence will be deployed across many economic sectors.
Amazon To Kill 600,000 Jobs And Use AI Instead (Secular Talk)
Seven important videos this week
The West Is Destroying Online Privacy (Vanessa Wingårdh)
Digital IDs Just Went Live — Say Goodbye To Your Privacy & Money (Tom Bilyeu)
It Begins: AI Brain Rot Is Destroying Your Brain (And It’s Getting Worse) (Moon)
The Warnings About Superintelligence Are Getting Louder (Novara Media)
WifeGPT: OpenAI Erotica Rollout + A Brief History of AI Companions (Lauranirion)
Three Charts That Help Explain What’s Behind the AI Bubble Fears (The Wall Street Journal)
Dotcom All Over Again: AI Shady Finance Deals (Breaking Points)
🔈 Listen
Why superhuman AI would kill us all
In the new episode of Modern Wisdom, podcaster Chris Williamson interviews Eliezer Yudkowsky, a researcher in artificial intelligence and the founder of the Machine Intelligence Research Institute.
They talk about Yudkowsky’s new book If Anyone Builds It, Everyone Dies. Why Superhuman AI Would Kill Us All, co-authored with Nates Soares. They talk about the existential risks associated with the creation of an artificial general intelligence.
Listen to Modern Wisdom on iHeart
Thank you for reading the 29th press review of Artificial reality. If you don’t have a paid subscription yet, you can support my work by clicking the button below, thanks!
Have a good week,
Arnaud
