Race to develop AI browsers

In the face of the AI ​​browser boom, experts warn of the risk of information leakage when attack code automatically activates as soon as users access the site normally.

OpenAI has just introduced Atlas, its first web browser. Integrated directly with ChatGPT, Atlas allows users to interact with artificial intelligence (AI) right in the web browser interface - from summarizing content, comparing data, quickly looking up to analyzing in-depth information.

User's personal assistant

Previously, Google launched an AI search feature in Chrome while Microsoft integrated Copilot into Edge. Start-ups like Perplexity with Comet browser or The Browser Company with Dia are also rushing to develop similar features. The common point of the new generation of browsers is to turn AI into a smart assistant for Internet users, instead of just an information search tool.

With Atlas, experts say the browser not only helps expand the ChatGPT ecosystem - which already has more than 800 million users/week - but also marks an important step for OpenAI when it officially steps into the field of smart search and browsing, where Google has dominated for 2 decades. However, it is unclear whether ChatGPT Atlas can compete with Google Chrome, a browser with more than 3 billion users worldwide.

On the user side, the appearance of AI-integrated browsers attracts a lot of attention, especially from office workers, researchers, and content creators.

Mr. Nguyen Van Tuan, an office worker in Ho Chi Minh City, said he used Atlas to search for consumer products and noticed a clear difference compared to the traditional search method. When he searched with the content "best phones in the price range of 7-10 million worth buying today", Atlas not only listed models such as Galaxy A35, Redmi Note 13 Pro or iPhone SE, but also provided user reviews, performance comparison, camera, battery... as well as analyzed strengths and weaknesses. "Atlas also suggests reasonable prices by region, whether to buy in installments or wait for further discounts. All are summarized in the same window, no need to open dozens of tabs to compare" - Mr. Tuan commented.

However, many users reflect that ChatGPT Atlas still has some limitations such as slow page loading speed when opening many tabs, small extension store and no cross-platform support like Chrome.

OpenAI officially launched the AI ​​browser ChatGPT Atlas on October 21, creating a fever among office workers and digital content creators.

Risk of data leakage

Mr. Tran Bao Dinh, General Director of TitKul Joint Stock Company, said that the emergence of AI browsers is changing the way people search for information. That is, switching from self-searching to direct interaction with AI to get quick content suggestions with higher personalization. However, this convenience leads to security risks as AI increasingly accesses users' personal data. Users should be cautious when searching for information through AI browsers as this type is becoming more and more popular.

According to experts, to operate smoothly, AI browsers often require deep access to users' emails, contacts, calendars and personal data - information that can be exploited in the event of a cyber attack. More dangerously, this form of malware attack is hidden in websites by inserting malicious commands to trick AI into performing unauthorized actions. When exploited, AI tools can accidentally disclose data, illegally access accounts or make transactions and posts without the user's knowledge.

Adrian Hia, Managing Director of Asia Pacific (APAC) of Kaspersky, said that in the first half of 2025 alone, Vietnam recorded about 300,000 cyber attacks, equivalent to 1,600 data leaks/day. The reason lies in the habit of sharing too much information and the lack of understanding of users. "The more data we provide to AI, the better the system understands us and the easier it is to exploit information if it falls into the wrong hands," Mr. Hia warned.

Mr. Hia cited a friend who used an AI platform to summarize internal financial reports. Shortly afterward, the system was discovered to have stored a copy of the data on a foreign server, and some of the content appeared publicly online. In another case, according to Mr. Hia, a passenger was denied boarding after his ticket was stolen because he shared an image on an open AI application.

According to research by Brave, an American technology company specializing in secure browsers, the risk of information leakage is a common problem with AI browsers. For example, with Atlas, shortly after its launch, researchers discovered a "clipboard injection" vulnerability, allowing malicious websites to automatically insert data into the clipboard. When users paste that content into a form or address bar, they can be redirected to a phishing site or automatically enter manipulated data.

What’s worrying is that this attack code isn’t pre-installed malware, but rather a hidden JavaScript code that activates itself when the user performs normal operations. Many malwares are even hidden in images or videos that contain hidden data, making it harder for AI systems to detect.

Experts recommend that users should not rush to grant AI comprehensive access, especially to bank accounts, medical records or internal documents. Use separate passwords for each service, enable two-factor authentication and limit sharing of photos, plane tickets, passports, family information via social networks or open AI platforms.

Source: https://nld.com.vn/chay-dua-phat-trien-trinh-duyet-ai-196251028210443257.htm