ChatGPT App Adds Computer Use and Built-in Browser

On July 16, 2026, OpenAI posted a demo video on its official X account showcasing two new capabilities for the ChatGPT desktop app: 'computer use,' which allows ChatGPT to directly operate other applications on the computer, and a built-in browser that lets ChatGPT browse the web, search for information, and navigate within the app. These two features are integrated into a single update, marking ChatGPT's transition from a purely conversational AI to an 'operational AI.'

The demo was hosted by OpenAI Developer Relations lead Dominik Kundel, who showed how ChatGPT can 'collaborate with your apps' and 'browse the web for research, navigate sites, and complete tasks.' Notably, OpenAI's wording emphasizes 'with you,' suggesting the feature is not fully autonomous but requires user participation or supervision.

Based on the demo, ChatGPT's 'computer use' feature likely relies on screen recognition and simulated click/keyboard input, similar to Anthropic's Claude Computer Use or Microsoft's Copilot Vision. This means ChatGPT can 'see' on-screen content, understand app interfaces, and perform actions like clicking and typing. The built-in browser may be based on Chromium or similar technology, enabling ChatGPT to access web pages, extract information, and fill out forms like a human.

However, publicly available information is limited. OpenAI has not disclosed the specific list of supported apps, whether the feature is available to all users, or if there are security restrictions (e.g., blocking access to banking or payment apps). Additionally, the demo did not show completion of multi-step complex tasks, such as cross-app workflows like 'organizing data in Excel and then sending a report via email.'

This update directly places OpenAI in the 'AI agent' competition. Previously, Anthropic's Claude launched Computer Use, and Microsoft's Copilot has similar capabilities, but ChatGPT's massive user base and brand recognition could accelerate adoption. If ChatGPT can reliably operate common office software (e.g., Office suite, browsers, email clients), it will upgrade from an 'information assistant' to a 'digital employee,' potentially disrupting the RPA (Robotic Process Automation) industry.

However, safety and reliability remain major challenges. AI operating a computer could make mistakes, leak sensitive information, or violate app terms of service. OpenAI's emphasis on 'working with you' in the demo suggests users still need to confirm key steps, which may be a risk mitigation strategy.

Currently, the feature has only been announced via a video on X; OpenAI has not released an official blog post or technical documentation. Key missing details include: whether the feature has been rolled out to all desktop users, whether a paid subscription is required, supported operating systems (Windows/macOS/Linux), and privacy measures (e.g., whether screenshots are uploaded to the cloud). Additionally, the demo did not show error handling scenarios—if ChatGPT cannot recognize an app interface or an operation fails, how will it respond?

Another open question is how the built-in browser and 'computer use' feature work together. For example, can ChatGPT search for information in the browser and then automatically fill data into a local app? This will be an important indicator of its automation level.