Log in Sign up
Opinion Mercor Mapped 29 AI Benchmarks Onto 18,797 American Job Tasks — and Found 90 Percent of US Working Hours Barely Measured Business Anthropic Quietly Pitched Its Own AI Safety Bill to Congress: Government Could Halt a Model Only if It Risks 50 Deaths or $1 Billion in Damages Claude In Under 30 Minutes Claude Solved a 26-Year-Old Cryptography Mystery: the NSA's Unexplained NIST Curve Seeds Came From lrand48 and Unix Timestamps ChatGPT ChatGPT Can Now Read and Act Inside Your Open Chrome Tabs — and It Works on Every Plan, Free Included News Reflection Ships Beam: a 501-Billion-Parameter Apache 2.0 Agent Model With 40-Plus Day-Zero Partners AI Agents Four Gemini Agents Beat Minecraft in 27 Hours — After 605 Deaths and 2,783 Self-Written Programs Business SoftBank Weighs a Takeover of Security Giant Allied Universal That Could Top $30 Billion — to Find Its Robots a Real Job Coding Assistants Replit's CEO Says the Company Has Dropped Anthropic for OpenAI — a Claim Insiders Dispute Opinion Mercor Mapped 29 AI Benchmarks Onto 18,797 American Job Tasks — and Found 90 Percent of US Working Hours Barely Measured Business Anthropic Quietly Pitched Its Own AI Safety Bill to Congress: Government Could Halt a Model Only if It Risks 50 Deaths or $1 Billion in Damages Claude In Under 30 Minutes Claude Solved a 26-Year-Old Cryptography Mystery: the NSA's Unexplained NIST Curve Seeds Came From lrand48 and Unix Timestamps ChatGPT ChatGPT Can Now Read and Act Inside Your Open Chrome Tabs — and It Works on Every Plan, Free Included News Reflection Ships Beam: a 501-Billion-Parameter Apache 2.0 Agent Model With 40-Plus Day-Zero Partners AI Agents Four Gemini Agents Beat Minecraft in 27 Hours — After 605 Deaths and 2,783 Self-Written Programs Business SoftBank Weighs a Takeover of Security Giant Allied Universal That Could Top $30 Billion — to Find Its Robots a Real Job Coding Assistants Replit's CEO Says the Company Has Dropped Anthropic for OpenAI — a Claim Insiders Dispute

ChatGPT Can Now Read and Act Inside Your Open Chrome Tabs — and It Works on Every Plan, Free Included

ChatGPT · By: AIReport Admin · 1 hour ago · 👁 7
ChatGPT Can Now Read and Act Inside Your Open Chrome Tabs — and It Works on Every Plan, Free Included

OpenAI upgraded its ChatGPT for Chrome extension into a full browser agent: it reads your open tabs, answers across several pages at once, and in Work mode can send messages, fill forms and organize tabs via Computer Use. The update, version 1.26.1005.11759, is available on every ChatGPT plan and in Chromium browsers including Edge, Brave and Opera.

OpenAI has quietly turned its ChatGPT extension for Google Chrome from a sidebar chat window into a full browser agent. The updated extension, version 1.26.1005.11759, can read the tabs you already have open, pull information from several of them at once, and — in its new Work mode — take actions inside those pages: sending messages, filling out forms, downloading files and grouping tabs. OpenAI says the upgrade works with every ChatGPT plan, including the free tier.

In chat mode, users can target specific tabs with an @chrome mention or simply ask a question that spans several pages. OpenAI's own examples read like shopping and research chores: comparing three sofa product pages by size, price, delivery time and return policy, summarizing a stack of articles in plain language, or matching up price pages from different sellers. The assistant now has the full chat history available in the sidebar, so follow-up questions do not require leaving the page you are on.

Work mode is the bigger change. It hands ChatGPT control of the tab through the same Computer Use capability that powers OpenAI's agentic products — GPT-6 Astra is currently the company's best model for screen-based work — and lets it operate inside sites where you are already logged in. OpenAI's demo flows include filling out a new dentist's intake form and adding the appointment to your calendar, or analyzing the terms of service of several contractors across tabs and drafting an email to the best fit.

That capability comes with a wide permission footprint. Installing or enabling the features can require the extension to read and change site data, access page debugging tools, manage tabs and downloads, and optionally read bookmarks and browsing history or show notifications. OpenAI gates each new website behind a consent prompt — allow once, always for this site, or always for all sites, the last of which the company itself flags as higher risk — and browsing history and file uploads or downloads need separate authorization.

The extension also works outside Chrome: OpenAI notes it runs in other Chromium-based browsers, including Microsoft Edge, Brave and Opera. The sidebar upgrade gives ChatGPT access to the user's full conversation history without navigating away from the current page, which makes the assistant feel closer to a persistent browser companion than an add-on.

The strategic backdrop is telling. OpenAI launched its own browser earlier this year, then shut it down on August 9, and has since poured its browser ambitions into the Chrome extension instead — meeting users where 3-plus billion Chrome users already are rather than asking them to switch. Rivals have taken the same route: Perplexity's Comet and The Browser Company's Arc both began as standalone products before shipping extension-based assist experience.

For users, the trade-off is straightforward: a browser assistant that can genuinely complete multi-step chores, in exchange for an agent that can see everything you have open. Security-minded reviewers of the update suggest closing sensitive logged-in tabs — banking, email, medical portals — before enabling Work mode, and relying on the per-site approval flow rather than the all-sites setting. The tab bar, it turns out, is now part of your AI attack surface.

Comments (0)

Log in to join the discussion

Log In

No comments yet