OpenAI has unveiled ChatGPT Atlas, a full-featured web browser designed around its conversational AI model, positioning ChatGPT as an ever-present assistant that integrates seamlessly into everyday browsing. The launch signals a shift toward agentic systems — intelligent browsers capable of performing complex tasks and learning from user context.

According to OpenAI, Atlas turns the browser into a workspace “where all your tools, work, and context come together,” eliminating the need to copy, paste, or switch tabs. Users can interact with ChatGPT in real time within any page, ask questions, generate summaries, or even automate actions directly in the same window.

Yogya Kalra, a college student and early tester, described how Atlas simplifies study workflows: “During lectures, I like using practice questions and real-world examples to really understand the material. I used to switch between my slides and ChatGPT, taking screenshots just to ask a question. Now ChatGPT instantly understands what I’m looking at, helping me improve my knowledge checks as I go."

With Atlas, users can opt in to browser memories, which allow ChatGPT to remember context from previously visited sites and reference it later. For instance, users can ask the assistant to “find all the job postings I was looking at last week” or “create a summary of industry trends so I can prepare for interviews" or "research holiday gifts based on the products I've viewed."

Browser memories are optional, private, and fully user-controlled. You can view all memories in settings, archive or delete them at any time, and clearing your browsing history also removes any related data. Even when browser memories are on, you decide which sites ChatGPT can or can’t access using the visibility toggle in the address bar. When visibility is turned off, ChatGPT cannot view page content, and no memories are created from it.

When enabled, ChatGPT can draw on these memories to offer personalized recommendations and smarter workflow assistance — from suggesting follow-up reading to creating lists based on recent browsing activity.

Earlier this year, OpenAI introduced ChatGPT agent, and now it’s been integrated natively and optimized for speed within Atlas. When using agent mode in Atlas, ChatGPT can perform tasks directly in the browser — opening tabs, navigating websites, and carrying out actions such as researching topics, booking appointments, or ordering groceries — all under user supervision.

For example, users planning a dinner party can share a recipe with ChatGPT, ask it to find nearby grocery stores, add ingredients to an online cart, and place the order for delivery. In professional settings, agent mode can open and review past team documents, perform competitive research, and compile insights into a comprehensive brief. When a request involves several steps, ChatGPT may confirm before opening new tabs or performing clicks, and users can also activate actions manually through the agent mode button in the toolbar.

Agent mode launches today in preview for Plus, Pro, and Business users. OpenAI emphasized that the feature remains an early experience, with ongoing work to improve reliability, latency, and complex task accuracy.

OpenAI said safety was a top priority in building Atlas, particularly as agentic systems gain the ability to take actions on logged-in sites. Agent mode is designed with layered safeguards:

Data privacy and parental oversight are built into Atlas. By default, OpenAI does not use the content users browse to train its models. Those who choose to opt in can enable “Include web browsing” in their data controls settings, which extends model training to chats conducted in Atlas, including content attached via the Ask ChatGPT sidebar and browser memories that inform those chats.

Parental controls also carry over automatically from ChatGPT to Atlas, ensuring consistency across experiences. New options allow parents to disable browser memories and agent mode, adding an additional layer of protection for families.

As OpenAI notes, agent mode remains an early experience and may make mistakes on complex workflows. The company is continuously improving reliability, latency, and success rates for multi-step tasks. Still, AI agents carry inherent risk. In rare cases, they may misinterpret commands or encounter hidden malicious instructions embedded in websites or emails — scripts that could override intended behavior and attempt to access sensitive data.

To mitigate this, OpenAI has conducted thousands of hours of red-teaming and built adaptive safeguards designed to respond quickly to new threats. However, no system can block every emerging attack as agentic AI grows more capable. Users are encouraged to balance convenience with caution, use logged-out mode when appropriate, and monitor agent activity during sensitive tasks. The company has pledged to continuously patch vulnerabilities as they are discovered. You can read more in their system card.

Inside Atlas, the ChatGPT sidebar can summarize pages, answer questions, or act on user commands without switching tabs. Users can:

The experience merges search, task automation, and AI writing assistance into one interface — a model that could redefine the role of browsers in everyday computing.

ChatGPT Atlas launches today for Free, Plus, Pro, and Go users on macOS, with Business and Enterprise availability in beta. Versions for Windows, iOS, and Android are in development.

Users can download the browser at chatgpt.com/atlas, import bookmarks and passwords, and start using ChatGPT natively within their browsing environment.

According to OpenAI, the Atlas roadmap includes multi-profile support, improved developer tools, and better app discoverability through the Apps SDK. Website owners can enhance agent compatibility by adding ARIA tags, and OpenAI plans to release frequent updates via its public release notes.

Q1: What is ChatGPT Atlas?
A: It’s a new web browser built by OpenAI that integrates ChatGPT directly into the browsing experience, enabling real-time AI assistance.

Q2: How does agent mode work?
A: Agent mode allows ChatGPT to take actions on websites — such as researching, booking, or shopping — within the browser, under user supervision.

Q3: Are browser memories private?
A: Yes. Browser memories are tied to your ChatGPT account, fully user-controlled, and not used for model training unless you opt in.

Q4: Which devices support ChatGPT Atlas?
A: Atlas is available now on macOS, with Windows, iOS, and Android versions coming soon.

Q5: Who can access agent mode?
A: Agent mode is in preview for Plus, Pro, and Business users, with enterprise support expanding later.

The launch of ChatGPT Atlas marks one of OpenAI’s most ambitious steps toward AI-native computing — where the browser itself becomes an intelligent, interactive workspace. By blending search, context awareness, and task automation, Atlas positions ChatGPT not just as a chat interface but as a true AI operating environment.

It also signals a major competitive moment in the evolving AI browser race, alongside products like Arc, Perplexity's Comet, and Brave AI Search. The integration of agent mode and browser memory establishes a new benchmark for contextual AI assistance — one that could transform how users learn, work, and manage daily digital life.

If successful, ChatGPT Atlas could redefine how people experience the web — shifting from passive browsing to active, intelligent collaboration. It suggests that the future of the internet may not just be powered by AI, but co-authored with it — becoming a partner in creation, learning, and decision-making across everything users do online.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant used for research and drafting. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.