ChatGPT can peek at your Chrome tabs now

The upgrade replaces the discontinued ChatGPT Atlas browser by granting the extension permission to scan any tab the user has open when they invoke “@chrome” inside ChatGPT’s web or desktop interface. In demo clips, the model pulls details from three sofa‑shopping pages, compiles a comparison spreadsheet, and even drafts an email to a contractor based on tabs showing service quotes. The sidebar version of ChatGPT now displays the full model with persistent conversation history, and “ChatGPT Work” can issue commands to Chrome such as sending messages, auto‑filling fields, or downloading files—functions that echo Google’s Gemini‑powered auto‑browse feature.

This move positions OpenAI squarely in the emerging AI‑assistant arms race where browsers become active platforms for large‑language‑model actions. Google has already embedded Gemini into Chrome, while Microsoft is layering Copilot across Edge and Office. By exposing tab contents, OpenAI is trying to make ChatGPT a more hands‑on productivity tool, blurring the line between a passive chatbot and an autonomous agent that can manipulate the user’s web environment. The feature also signals OpenAI’s strategy to keep its ecosystem sticky after pulling the Atlas desktop app, leveraging the ubiquity of Chrome to stay in users’ daily workflows.

The ability to read every open tab raises immediate privacy and security questions. Users must trust that OpenAI’s extension respects the same‑origin policy and does not exfiltrate sensitive data, especially when the model can draft emails or submit forms on the user’s behalf. Regulators may scrutinize consent mechanisms, and enterprises could block the extension on corporate devices. Watch for OpenAI’s forthcoming policy disclosures, potential granular permission controls, and whether competitors respond with tighter sandboxing or additional user‑control features.