OpenAI has announced 'ChatGPT Atlas', a new browser that allows users to interact with artificial intelligence (AI) while web browsing. It was released for Mac users on the 21st (local time), with Windows, iOS, and Android versions scheduled to be released in the near future.

While OpenAI added internet-based answering capabilities to ChatGPT last year, it has emphasized the importance of the browser, where a user's tasks, tools, and context are all concentrated. The company stated that it will provide ChatGPT Atlas as a true 'super assistant' that understands the world and supports goal achievement.

When you launch ChatGPT Atlas, the familiar ChatGPT logo and a text input field appear, allowing you to ask questions and use it just like the existing ChatGPT. Additionally, by selecting the 'Search Links', 'Images', 'Videos', and 'News' tabs at the top of the screen, you can check specific search results beyond chatbot conversations. Users can also display and use a separate sidebar for chatting with the bot while browsing web pages.

This browser is equipped with a 'Browser Memory' feature that remembers user actions and allows for easy retrieval of information from search history. For example, if you command the chatbot to "show me the shoes I looked at yesterday," ChatGPT will identify and present the website containing that information from your history. Furthermore, it includes features such as the AI creating a To-Do list based on recent activity or investigating and suggesting relevant sale information when a user views a specific product.

The Browser Memory feature is exclusive to ChatGPT accounts and is disabled by default for privacy protection. Unless the user manually enables it, ChatGPT does not view page content or create memories. Even when memory is enabled, browser memory and history management are possible through the settings menu. Additionally, users can switch the browser to Incognito mode and use it while logged out of ChatGPT.

Regarding data training, ChatGPT Atlas does not use content browsed by the user for AI model training in its default settings. However, if the user enables 'Include Web Browsing' in the data management settings (opt-in), the browsed pages will be used for AI training. If chat content training is enabled in the ChatGPT account, it also applies to chatbot conversations in the Atlas browser. In this case, attached website content or browser memory provided to the chatbot while using the sidebar is also included in the training data.

ChatGPT Atlas also features an Agent Mode utilizing the 'Operator' technology announced by OpenAI earlier this year. This feature allows ChatGPT to browse the web and execute specific tasks on behalf of the user. Agent Mode is currently in preview and is available on a trial basis only for ChatGPT Plus, Pro, and Business account holders.

In addition, a Parental Control feature has been added. If a guardian sets this in their ChatGPT account, the control also applies to chatbot conversations through the Atlas browser. Options for guardians to individually disable Atlas's browser memory and Agent Mode have also been added.

Recently, AI companies have been actively entering the AI-integrated browser market. Perplexity has launched the AI browser 'Comet', and Google is also integrating its Gemini AI assistant into the Chrome browser, intensifying competition in this field.