OpenAI Lets Existing ChatGPT Projects Switch Memory Modes
Eligible unshared projects can now move between default and project-only memory, while shared projects stay isolated and ChatGPT Work remains unavailable in project-only mode.
OpenAI updated ChatGPT Projects on August 14 so eligible existing projects can switch between default memory and project-only memory after creation. The change applies across ChatGPT plans, but shared projects remain locked to project-only memory and ChatGPT Work is not available inside that mode.
The practical change is control after setup. A user no longer has to treat the original memory choice as permanent for every eligible unshared project. That matters for projects whose sensitivity, collaborators or purpose change over time, although OpenAI says a settings change can take a few hours to take effect.
What changed in project settings
OpenAI says users can open an existing project, select its three-dot menu, enter Project settings and choose either `Default memory` or `Project-only memory` under Memory. The update removes the need to create a replacement project merely to choose a different memory mode.
The availability statement is broad: OpenAI lists the change for all ChatGPT plans. The important qualifier is the project itself. Shared projects continue to use project-only memory and cannot switch to default memory. OpenAI does not define every other eligibility condition in the changelog, so users should not assume that every project will expose both choices immediately.
What project-only memory keeps separate
OpenAI describes project-only memory as a contained context boundary. ChatGPT can use conversations from the same project, but it does not reference memories or conversations from outside that project. Information from the project is also excluded from memory used in chats outside the project.
That is narrower than simply organizing conversations in a folder. The mode controls which prior context ChatGPT may draw on and where information from the project may influence later chats. Switching an eligible project therefore changes the context boundary, not only the way the project appears in the interface.
The source does not claim that project-only memory is a separate security product, a compliance control or a substitute for workspace data governance. It explains context behavior. Teams handling sensitive material still need to use their plan’s access, retention and administrative controls rather than treating one memory setting as a complete protection layer.
The remaining limits
Two restrictions are explicit. Shared projects stay on project-only memory, so collaborators cannot move a shared workspace into default memory. ChatGPT Work is also unavailable in projects that use project-only memory. Users who want Work inside a project therefore face a current feature trade-off between that workflow and the tighter project-specific context boundary.
There is also a timing caveat: OpenAI says changes may take a few hours to take effect. A user testing the boundary immediately after switching modes could therefore see a transition period rather than an instant, verifiable change. The changelog does not publish rollout measurements, independent privacy testing or a technical audit of memory isolation.
Why this is a meaningful product delta
Project settings often outlive the assumptions made when a workspace is created. A research project may later include confidential source material; a personal project may become shared; or a team may decide that cross-chat recall is useful only for part of its work. Post-creation switching makes those decisions reversible for eligible unshared projects.
The clearest next checkpoints are broader visibility of the setting, support for ChatGPT Work in project-only mode, and any future option for administrators to govern which memory mode a project may use. Until OpenAI documents those changes, the confirmed scope is limited to the controls and exclusions in the August 14 release note.
Status
Confirmed. Internal confidence is medium because the feature and its stated boundaries come from OpenAI’s official product changelog, but there is no independent rollout test or technical validation of the memory boundary.
Sources
Update note: Last reviewed 2026-08-17. We will revise this post if OpenAI changes project eligibility, shared-project behavior, propagation timing or ChatGPT Work support.
Sources
- OpenAI — ChatGPT release notes — official
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.