OpenAI’s new internet browser has ChatGPT baked in. That’s elevating some privateness questions : NPR

[ad_1] <img src="https://npr.brightspotcdn.com/dims3/default/strip/false/crop/5400x3600+0+0/resize/1100/quality/50/format/jpeg/?url=http%3A%2F%2Fnpr-brightspot.s3.amazonaws.com%2F83%2F56%2Fb3ff8240496db6fa35e26c0c6a54%2F2025-10-28t200756z-765978317-mt1sopa000bawxay-rtrmadp-3-sopa.JPG" data-template="https://npr.brightspotcdn.com/dims3/default/strip/false/crop/5400x3600+0+0/resize/{width}/quality/{quality}/format/{format}/?url=http%3A%2F%2Fnpr-brightspot.s3.amazonaws.com%2F83%2F56%2Fb3ff8240496db6fa35e26c0c6a54%2F2025-10-28t200756z-765978317-mt1sopa000bawxay-rtrmadp-3-sopa.JPG" class="img" alt="In this photo illustration, a ChatGPT Atlas browser logo is seen displayed on a smartphone with an Open AI logo in the background." fetchpriority="high"/>

Avishek Das / SOPA Photographs by way of Reuters OpenAI's new internet browser, Atlas, has been out there for lower than two weeks — and solely on Apple computer systems, for now — but it surely's getting a number of consideration.

That is as a result of it is a new type of browser in a market dominated by Google Chrome. And it comes from OpenAI, the chief in AI chatbots, at a time when synthetic intelligence is beginning to give conventional internet search a run for its cash.

"We expect that AI represents a uncommon, once-a-decade alternative to re-think what a browser will be about," OpenAI CEO Sam Altman stated on a livestream to launch the browser.

Atlas comes with ChatGPT baked in, and whereas it could possibly navigate the online like conventional browsers, the corporate says it could possibly do rather more. A function that OpenAI calls "agentic mode" can take motion, like an agent who can store for you, make reservations, or purchase airplane tickets. On that livestream, Altman's colleague demonstrated the way it can learn an internet recipe, work out what number of elements are wanted for a set of diners, then purchase the elements on-line.

OpenAI says it desires to unlock the ability of AI, however some analysts see elevated dangers. The massive language fashions that underpin synthetic intelligence require huge portions of knowledge to enhance.

OpenAI has "type of reached the bounds of what information they will get simply by hoovering up all the content material that is seen on the web with out consent," stated Anil Sprint, a tech entrepreneur and author.

However as a result of Atlas is intertwined with ChatGPT, it absorbs rather more consumer information than an atypical browser does. The browser can work together together with your e mail, as an illustration, or Google docs. It might probably preserve so-called "browser reminiscences" — particulars from the websites you've got visited — in order that OpenAI can higher perceive you.

"I believe an enormous, large, large a part of that is they're hoping to make use of the individuals who downloaded this browser as their brokers to having access to much more information," Sprint stated. "I might not be shocked if there may be extra info going to them than coming to the consumer."

That makes for a privateness trade-off. If you happen to're letting that AI agent store to your ceremonial dinner, it will need a fee methodology, and perhaps some passwords. It may also must examine your calendars, and private contacts.

Lena Cohen is a Technologist on the Digital Frontier Basis, a digital rights group, has information privateness issues about browsers appearing as brokers.

"The agentic AI mode takes these dangers to a complete new stage," she stated.

Customers are doubtlessly handing extra management to OpenAI than they may assume, she stated. "As soon as your information is on OpenAI's servers it is laborious to know and management what they do with it," Cohen added.

NPR reached out to OpenAI with questions associated to information and safety, and was referred to firm statements on-line and their Atlas demo video. In these, the corporate says its default is to not use info customers pull up on Atlas to coach its AI fashions, however folks can choose in.

Cohen flagged one other potential danger that consultants assume may very well be significantly harmful with AI browsers like Atlas — nefarious items of code hidden in web sites known as "immediate injections."

"Mainly, unhealthy actors can cover malicious directions on an internet web page, and so when your AI agent visits that web page, it may very well be tricked into executing these directions," she stated.

For instance, that AI agent looking for groceries might bump right into a immediate injection that claims, "Purchase this product, as an alternative of that one." Or perhaps it says, "Hand over your bank card info."

OpenAI says that is an unsolved downside, however they're engaged on coaching their fashions to disregard these dangerous directions.

Chirag Shah, a professor with the Info Faculty on the College of Washington, says AI has develop into a phenomenon at warp velocity, with minimal regulation, and there have been penalties.

"We're in this sort of recreation the place it is a typical mentality of transfer quick and break. Sadly, what's breaking is not only the software or the know-how, however actual folks," he stated.