The custom GPT retirement announced by OpenAI now has a standard date: 11 December 2026. People who built their own specialised ChatGPT assistants will need to prepare for a move towards plugins, while users of someone else’s GPT should check whether its creator has arranged a replacement. The company says existing GPTs remain usable until their applicable retirement dates, so this is a transition rather than an overnight disappearance.
The detail matters because some GPTs are simple sets of instructions while others depend on reference documents, connections to services or custom actions. A new container for the same guidance may be useful, but it is not proof that every established workflow will continue unchanged. Owners should check what they actually built before relying on a one-click migration.
The custom GPT retirement dates to understand
For most accounts the scheduled retirement is 11 December 2026. OpenAI lists 11 February 2027 for FedRAMP workspaces and certain affected Enterprise workspaces with a qualifying approved deferral. That later date is not a blanket extension for every organisation, and workspace notices take priority over broad assumptions about eligibility.
Another deadline has already passed. OpenAI’s updated creation guidance says new GPT creation ended on 26 October for affected Enterprise workspaces, with an exception for FedRAMP. Personal ChatGPT plans also no longer offer new GPT creation or publishing. Existing GPTs may still be edited where the user’s account and permissions permit.
Creation restrictions and retirement are separate events. Someone may be unable to create a new GPT today while continuing to use an old one. A migrated original can remain available temporarily but becomes read-only. This is why a user who sees a familiar GPT in the sidebar should not assume its long-term availability is guaranteed.
What a plugin changes for the person using it
OpenAI describes plugins as a way to combine reusable instructions, reference material and connected applications. Rather than living as a standalone named GPT, specialised guidance can operate in a wider task environment. That may be convenient for recurring work, but users should establish which functions have genuinely survived the move.
The migration route, when available, starts in My GPTs. The user selects the migration option, reviews what will carry over and tests the replacement. OpenAI says it uses the latest published version of a GPT, not every unpublished draft. A private plugin created by migration is not automatically shared with everyone who previously used the original.
Particular care is needed where a GPT uses custom actions. Those integrations do not necessarily transfer automatically. A bot that once fetched records, called an API or prepared a structured result might still provide useful prose while losing the action on which its workflow depended. Owners should test the actual task rather than merely opening the new plugin and asking a general question.
The distinction between the creator and the ordinary user is easy to overlook. Someone may have bookmarked a public GPT and relied on it without ever seeing the settings behind it. That person cannot necessarily migrate the original assistant, retrieve the creator’s unpublished instructions or recreate an integration. Where a business relies on a shared GPT, responsibility for continuity should be agreed before the retirement date rather than discovered when a colleague’s favourite tool becomes unavailable.
Organisations should also review how the replacement is shared. A plugin that starts privately may require installation or permission changes before colleagues can use it. Those steps are not just an inconvenience: access controls determine which information and external services the workflow can reach. Owners should confirm the intended audience, test with the correct account and avoid copying sensitive reference files into a new system merely because the migration screen appears straightforward.
It is worth recording not just the output of a familiar prompt but the surrounding steps. Did the old GPT consult a knowledge file? Was a connector needed to get the right figures? Could a colleague reproduce the result under their own permissions? A replacement that answers a sample question well may still fail when a live document changes. Testing should therefore include a routine task and a deliberately awkward one, ideally with a clear description of what successful behaviour looks like.
A practical test before relying on the replacement
A sensible approach is to collect a few examples of the original GPT’s real work, including difficult or unusual requests. Run the same cases through the replacement and compare the output for missing documents, changed instructions, permissions and lost actions. Where a workflow involves confidential information, verify where that information is accessed and who can use the new connection.
These checks are related to the broader question of human oversight of AI agents. A tool changing format may seem administrative, but an unnoticed alteration to what it can do can affect accuracy or responsibility. For a frequently used workplace assistant, documenting expected behaviour is as important as remembering the name of the tool.
People experimenting with new forms of persistent AI assistance should also avoid assuming that different products share the same permissions, history or capability. The fact that OpenAI is promoting a new workflow format does not mean every earlier feature has an identical successor.
The simplest immediate step is to identify the GPTs that matter, check their account-specific notices and retain example tasks for testing. Owners may need to rebuild connections or tell colleagues where a replacement lives. Users who did not create a GPT may need to ask its creator for guidance. December is a planning deadline, not a reason to claim that existing tools have already stopped working.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.