AI Browser

An AI browser, or agentic browser, is a web browser that integrates a generative artificial intelligence assistant as a core feature rather than an add-on. Users can ask questions about the page they are reading and, in "agent" modes, delegate multi-step tasks — research, comparison shopping, form-filling, booking — to AI that acts directly within the browser. The category emerged in 2024 and 2025 as AI companies sought a position in how people access the web, and drew attention again in 2026 when OpenAI retired its standalone entry, ChatGPT Atlas.^[1][2][3]

History

Early precursors included AI features added to established browsers, such as Microsoft's Copilot integration in Edge, and extensions that added chatbot sidebars. The Browser Company's Arc introduced a "browse for me" feature in 2024, and Perplexity — an AI search company — released Comet in 2025, one of the first browsers designed around an assistant, with an emphasis on research and task automation.^[1][7]

OpenAI launched ChatGPT Atlas, a standalone desktop browser with ChatGPT built in, on 21 October 2025, initially for macOS. It combined a chat sidebar that understood the current page with "browser memories" and an agent mode that could complete tasks such as ordering groceries or assembling team briefs. Reviewers praised its integration but questioned its reliability and privacy practices, noting that page content is processed on the vendor's servers.^[2][5]

The standalone-browser strategy proved short-lived, however. In July 2026, OpenAI announced that Atlas would be discontinued, and it stopped working on 9 August 2026, with the company redirecting users to agentic features inside the ChatGPT desktop app and a ChatGPT extension for Chrome — arguing it could bring agentic browsing to users where they already are. By late 2026, the category's centre of gravity had shifted from dedicated browsers towards AI features inside the browsers people already use: Comet, Google Chrome with Gemini and Microsoft Edge with Copilot Mode all continued to develop their assistant and agent features during the year.^[3][4][6]

How they work

AI browsers differ from conventional browsers in two ways. First, an always-available assistant can see the page being viewed — reading text, images or page structure — so users can ask questions without copying content elsewhere. Second, agent mode lets the AI navigate, click and type across sites to finish tasks, using the same underlying models that power general AI agents.

This design concentrates data in the browser: page contents, browsing context and sometimes history are sent to the vendor's models for processing rather than staying on the device, and several products offer opt-outs or logged-out modes for sensitive tasks. Because agents can act inside logged-in sites, security researchers warn about prompt injection — malicious instructions hidden in web pages that can hijack an agent's actions — alongside risks of automated misuse in fraud and shopping scenarios.^[1][2]

>See Also

• Perplexity AI
• Prompt Injection
• AI Agents
• Chatbot
• Semantic Search
• PDPA and AI Compliance