Custom GPTs move to plugins before 11 December, and custom actions need rebuilding

On 26 October 2026, OpenAI turned off new custom GPT creation for affected ChatGPT Enterprise workspaces. FedRAMP workspaces are the exception. Existing GPTs that haven't been migrated stay usable and editable until their retirement date. For new reusable workflows, OpenAI points people to Plugin Creator.

OpenAI's FAQ says custom GPTs are scheduled to retire on 11 December 2026. On that date, custom GPTs and their GPT pages become inaccessible. FedRAMP workspaces, and affected Enterprise workspaces with a qualifying, approved deferral, retire on 11 February 2027 instead.

OpenAI says the later date doesn't apply to every workspace. A deferral also doesn't keep the ability to create new GPTs. Existing conversations with custom GPTs stay accessible after retirement, with no action needed to keep them.

The change affects all ChatGPT plans. OpenAI plans to help creators move their GPTs to plugins, which bring reusable instructions, reference files and connected apps together. Migration and plugin access may differ by account or workspace.

For Enterprise, the FAQ lists the steps so far. Admins got a notice on 11 September 2026. OpenAI's target for the migration tool and a user banner was 22 September, or 1 October for workspaces that chose delayed banners. OpenAI calls both dates targets.

Standard admin emails carry the standard dates and aren't tailored to workspaces with an approved delay. For anyone who uses several accounts or workspaces, the timeline and permissions of the one where the GPT was created apply.

By our arithmetic, 26 October to 11 December is 46 days, about six and a half weeks. Every custom GPT a team relies on has to be migrated, tested and shared again in that time, unless the workspace has an approved deferral.

Teams that want a new reusable workflow now build it as a plugin. An existing GPT can still be edited until it is migrated, so fixes can go in first and travel with it.

An admin email with the standard dates doesn't tell you whether your workspace has a deferral. OpenAI points admins to the workspace notice or the account team for that.

Staff keep their old conversations. The GPT pages go, so links to them from intranet pages and how-to guides will likely stop working after 11 December.

A GPT's creator migrates it from My GPTs with "Migrate to plugin", and in an Enterprise workspace a workspace admin can migrate it too. Plugins have to be enabled and the GPT has to be published, which doesn't mean sharing it publicly.

The GPT's instructions become a skill inside the new plugin. Its connected apps are added to the plugin as apps, and its knowledge files are copied into the plugin's reference files.

The GPT's selected model doesn't carry over, and in Enterprise workspaces the Enterprise defaults apply. After migration, the original GPT stays usable until retirement. It is read-only, and its creator can't delete it.

Custom actions, which let a GPT call another system, don't transfer. OpenAI says features that depend on them won't work in the plugin until a replacement is set up. Rebuilding the connection may need a custom MCP server, a service that connects the plugin to the other system, plus technical setup.

OpenAI says a rebuilt integration shouldn't be assumed to do everything the original action did, and advises extra time for GPTs that use custom actions. Access to custom integrations depends on the account or workspace.

In Enterprise, a plugin migrated on its own starts private, and the GPT's users don't get access automatically. Share Plugin can fill in the original GPT's users, groups or workspace audience, for someone to review and save.

Sharing with people or groups needs the "Share plugins" permission. Publishing to the workspace directory needs "Publish plugins to workspace", a separate permission. To use a migrated plugin, a person needs "Use plugins" and a workspace policy that lets them install it or installs it for them.

Staff call a plugin or skill with an @ mention, or through + then More. ChatGPT may also use an installed skill on its own, but OpenAI says the plugin won't necessarily run on every request.

Bulk migration runs through the ChatGPT Admin plugin, in the background, and starting it doesn't confirm that every GPT migrated. Public GPTs made in affected Enterprise workspaces are included even when people reach them from a personal account. Their public sharing settings don't carry over.

On cost, OpenAI says creating and editing plugins through a conversation uses Work or Codex, and actions in Work consume credits.

For a GPT built from instructions and files, the move is mostly a click and a review. GPTs with custom actions take engineering time, because their features stop working in the plugin until someone rebuilds the connection.

A GPT that reads from or writes to an internal system through a custom action is likely a small build of its own. Plan it with its own permissions, testing and security review, and allow the extra time OpenAI advises.

Because the plugin starts private, staff who used the GPT won't see it until someone shares it. If nobody does before 11 December, they lose the tool when the GPT retires.

A GPT set to one model now runs on the Enterprise default, and OpenAI says the plugin may respond differently. Each plugin that a team depends on is worth testing before the old GPT goes.