OpenAI has introduced Atlas, a web browser powered by ChatGPT, aiming to transform how users navigate the internet. Released on Tuesday for macOS, this innovative tool allows users to direct ChatGPT through text or voice commands, enhancing web interactions while simultaneously training the AI.
In a livestream event, OpenAI CEO Sam Altman emphasized that Atlas offers a chance to “rethink what a browser can be about.” This reimagining involves ChatGPT observing and guiding users through their browsing sessions. “With Atlas, ChatGPT can accompany you anywhere on the web, understanding your intentions and assisting in completing tasks directly on the page,” OpenAI stated in a blog post.
According to the release notes, Atlas features ChatGPT in a sidebar, enabling it to analyze and summarize webpage content. Users can invoke ChatGPT from form fields on pages rendered by Atlas. The browser also enables users to manage their browsing history, allowing them to restrict ChatGPT”s access to specific sites. Pro and Business users can utilize an agent mode, which allows Atlas to perform more complex, multi-step functions. In this mode, ChatGPT can handle tasks such as researching meal plans, compiling shopping lists, and placing grocery orders for delivery, all while ensuring users are informed before significant actions are taken.
OpenAI”s strategy parallels that of competitors like Perplexity”s Comet browser, which debuted recently. The goal is to encourage users to delegate their online interactions to AI agents that can act on their behalf, including making purchases. Furthermore, OpenAI aims to expand ChatGPT”s reach, as web browsers serve as primary gateways for internet access. Major players like Google and Microsoft have already integrated AI services into their browsers, with Google incorporating Gemini into Chrome and Microsoft adding Copilot to Edge. Smaller companies like Brave and Opera are also following suit, while Mozilla now offers multiple AI models in Firefox.
Despite these efforts, it remains unclear whether Atlas will significantly alter the competitive landscape of AI in web browsing. A recent report from Menlo Ventures indicates that approximately a quarter of Americans have used either ChatGPT or Google Gemini recently, highlighting that consumer adoption is heavily influenced by embedded products with established distribution.
OpenAI”s ownership of a browser could also be strategically beneficial if it decides to venture into online advertising, an area that Altman previously dismissed but seems increasingly plausible as OpenAI faces substantial financial losses. The Atlas browser has some peculiarities; for instance, Developer Tools can only be accessed via the View menu but not on a New Tab page, likely due to the default ChatGPT input not functioning as a traditional webpage. Additionally, suggested prompts reappear in the New Tab window after being dismissed.
Built on Google”s Chromium open-source project, Atlas supports Chrome extensions. However, unlike Chrome, which currently allows users to disable AI Mode, there is no equivalent option in Atlas. OpenAI appears to prefer that users set Atlas as their default browser, even offering increased ChatGPT usage limits for those who maintain it as their default for a week.
The launch of Atlas raises important privacy considerations. Currently, there is no dedicated Privacy menu in the settings. The release notes assert that user browsing content will not be utilized to train AI models by default. However, upon installation, the Data Controls menu indicated that the default setting was “On,” meaning user content could be used for model improvement.
Pranav Vishnu, product lead for ChatGPT Atlas, acknowledged the potential risks associated with sharing browsing data. He stated that while Atlas offers powerful capabilities, it also introduces new risks. To mitigate these, Atlas operates solely on user tabs and cannot execute code or access local files on the user”s device. Concerns about security and privacy have been echoed by developers and users alike, with many expressing hesitation to trust AI-driven browsers until they undergo rigorous independent testing.
