ChatGPT → Codex/Work handoff approval card is too easy to approve accidentally

When ChatGPT decides that a task should be handed off to Codex/Work, it presents an approval card in the conversation.

The current interaction makes that handoff unusually easy to approve unintentionally. In our case, normal interaction with the conversation resulted in the handoff being approved when that was not the intended action. Because approval can initiate a materially different workflow, an accidental click/tap has more consequence than an ordinary chat UI interaction.

Expected behavior

The handoff should require a clear, deliberate approval action, and ordinary interaction with the conversation/card should not be interpreted as approval.

Actual behavior

The approval card can be activated too easily during normal interaction, making it possible to enter Codex/Work without intentionally choosing to hand the task off.

Why this matters

This is more than a navigation annoyance: the approval represents consent to move the task into a different execution environment. The UI should make the boundary between viewing/interacting with the handoff suggestion and approving the handoff unambiguous.

A dedicated approval control with a sufficiently constrained click/tap target, rather than making a broader card interaction equivalent to approval, would reduce accidental handoffs.

Start by locating the ChatGPT-to-Codex/Work handoff approval card and reproduce the accidental approval during ordinary conversation interaction. Trace which card area activates the handoff and verify that viewing or interacting with the card does not approve it. Done means approval requires a clear, dedicated action with a constrained click/tap target.