Hello OpenAI team,
I would like to propose a feature idea for ChatGPT that could significantly improve real-world web interaction.
Idea:
ChatGPT could create a sandbox environment inside a Chrome browser tab, where the model can observe the page (via screenshot or DOM snapshot) and propose actions such as:
- scroll
- click
- open link
- type text
- wait
The key part would be user approval before execution.
Example:
AI proposes → “Click the ‘Next’ button”
User sees the proposed action → Approve / Deny
If approved → ChatGPT executes the action inside the browser tab.
UI concept:
ChatGPT
└ Chrome Browser Tab Sandbox
├ Screenshot / DOM view
├ Proposed action timeline
├ Click / Scroll / Open link actions
└ Approve / Deny controls
This could also work through a ChatGPT Chrome Extension, where each Chrome tab becomes a sandbox environment that ChatGPT can safely interact with.
Benefits:
- safer AI interaction with websites
- user-controlled execution
- practical workflows for browsing, research, and automation
- Chrome tab-based sandboxing for safety and isolation
I believe integrating a sandboxed Chrome browser interaction system directly into ChatGPT would make it significantly more useful for real-world tasks and web workflows.
As an individual developer, I would also be very interested in experimenting with this capability if it ever becomes available through the ChatGPT platform or API.
Thank you for considering this idea.
Real workflow example:
In my own testing with browser-based interactive tools, I found that this interaction model can be extremely powerful for real development workflows.
For example, I was able to let the model repeatedly inspect outputs, run checks, and iterate until it eventually found and fixed a bug in an RTL hardware design that had previously gone unnoticed.
This kind of iterative debugging loop could make ChatGPT much more useful for developers if similar browser sandbox capabilities were available.