ChatGPT’s irreversible Project Memory setting is a serious problem for professional users

I recently discovered that one of the most important settings in ChatGPT Projects cannot be changed after a project has been created. When creating a project, users can choose between standard memory and project-only memory. However, if a project was created with standard memory, it apparently cannot later be converted to project-only memory. The official workaround is to create a new project and manually move the relevant chats into it. Project sources and instructions may also need to be added again. This is not just inconvenient. It creates a serious problem for anyone who uses ChatGPT Projects to separate clients, companies, legal matters, technical work, or other confidential contexts. With standard memory, a project is not necessarily isolated from the rest of the account. Information from conversations outside the project may be used inside it, while information from the project may potentially influence other chats. That defeats one of the main reasons for using separate projects. This has already been discussed on Reddit Several Reddit users have reported related problems. A discussion titled **“Project-only memory is finally here”** pointed out that project-only memory can only be selected when a new project is created. Users noted that existing projects cannot simply be converted and that the practical workaround is to create a new project and move the chats manually. Another thread, titled **“Can’t edit the settings when creating a new project?”**, included users who were unable to change the memory configuration after creating their projects. Other discussions have raised concerns about projects referencing memories or information from outside their intended project context. Users have also reported cases in which information from project chats appeared to influence global memory or other conversations. There are even anecdotal reports claiming that projects configured with project-only memory have still appeared to recognize information from unrelated chats. These reports are not official confirmation of a technical failure, but they show that users do not feel they can reliably verify whether project isolation is actually working. More recently, users have also reported restrictions when moving chats into or out of projects configured with project-only memory. If this behaviour is intentional, it makes fixing an incorrectly configured project even more difficult. What OpenAI officially confirms OpenAI’s own documentation confirms several important limitations: * Project-only memory must be selected when a new project is created. * An existing project using standard memory cannot later be changed to project-only memory. * There is no global setting that automatically makes all new projects project-only. * Users who want project-only memory for an existing project must create a new project and move their conversations. These consequences are significant, but they are not explained clearly enough when a project is created. Why this matters for professional users I manage different projects specifically to keep clients and subjects separate. I do not want information from one customer project influencing answers in another project. I also do not want unrelated personal or professional conversations being used inside a client project. For agencies, developers, consultants, lawyers, researchers, and businesses, context separation is not an optional convenience. It is a basic requirement. A project should either be isolated or clearly labelled as not isolated. The current design creates several problems: A consequential memory decision is made during project creation. The choice cannot be changed later. The consequences are not explained prominently enough. There is no account-wide default for isolated projects. Repairing the mistake requires manual migration. Users cannot easily verify which external memories influenced a response. Moving chats and sources between projects is unnecessarily cumbersome. ## What should change OpenAI should provide: * The ability to convert an existing project to project-only memory. * A global default setting for all newly created projects. * A clear warning before creating a project with standard memory. * A visible indicator showing whether a project is isolated. * A bulk migration option for chats, files, sources, and project instructions. * Clear information about which memories or conversations influenced a response. * Reliable guarantees that project-only memory does not access outside context. The warning should say something like: **“Standard memory does not isolate this project. Conversations and saved memories outside this project may influence responses. This setting cannot currently be changed after project creation.”** That would at least allow users to make an informed decision. As things stand, this is not merely a missing convenience feature. It is a serious transparency and project-separation problem, especially for professional users handling multiple clients or sensitive contexts.