Reported by @grand prawn
Bug Report: Cross-chat context/code mixing in GPT-5.2
`Steps to Reproduce`
Start two separate GPT-5.2 chats, each with different project plans and constraints.
Work in parallel for multiple sessions/days, switching back and forth.
Observe that after enough iterations, the model begins referencing the other project’s rules, files, or code in the wrong chat.
`Expected Result`
Strict isolation between conversations/projects
a warning when the model is uncertain about which project context it’s using
“pinned workspace/project” mode that prevents cross-project drift
Better recovery tools (rollback/restore/branch-safe workflows)
`Actual Result`
The model began mixing instructions, code, and plans between the chats.
`Environment`
macOS, Google Chrome, ChatGPT GPT-5.2 and App