Custom GPTs Retire on December 11, 2026: A Migration Plan for Private Equity and Credit Firms

1. The Short Answer: Count, Then Migrate, Move or Retire

OpenAI has scheduled custom GPTs to retire on December 11, 2026, according to its custom GPT retirement and migration FAQ. FedRAMP workspaces and Enterprise workspaces with an approved deferral get until February 11, 2027. On the retirement date, custom GPTs and their pages become inaccessible, while past conversations "will remain accessible after retirement."

So every GPT your firm relies on needs a new home or an end date. The sensible order: count them first, migrate the few that people use, move or rebuild the ones that touch deal systems, and retire the rest.

Migrating to a ChatGPT plugin is the easy route. Moving a GPT to Claude or Microsoft 365 Copilot, or giving it a proper build, is the bigger choice, and it suits firms already heading that way.

GPTs were easy to make, so expect the list to be longer than anyone remembers. The retirement forces the count nobody had time for.

2. What OpenAI Announced, and Which Date Is Yours

OpenAI announced the change in its ChatGPT release notes on September 11, 2026: "We're planning to retire custom GPTs across ChatGPT plans and provide a migration path to plugins, which can combine reusable instructions with connected apps."

The FAQ sets out the dates by workspace:

• Most workspaces. "Custom GPTs are scheduled to retire on Dec 11, 2026."
• FedRAMP and deferred Enterprise workspaces. FedRAMP workspaces and affected Enterprise workspaces "with a qualifying, approved deferral have a retirement date of Feb 11, 2027." A deferral does not preserve the right to create new GPTs.
• New GPTs in Enterprise. "Oct 26, 2026: New GPT creation ended for affected Enterprise workspaces, with the exception of FedRAMP workspaces."
• Personal accounts. "New GPT creation and publishing are not available on personal ChatGPT accounts, including Free, Go, Plus, and Pro," says OpenAI's GPTs article.

One line in the FAQ decides which date is yours: "What matters is where the GPT was created, not only your own plan."

Dates can still move. OpenAI's release note says "Dates may change," and the FAQ had been edited on the day it was read for this guide, October 28, 2026. Confirm the date in your own workspace notices.

Plan for December 11. If an extension arrives, treat it as spare time.

3. What Migrates and What Stays Behind

OpenAI's FAQ is specific about the mechanics, and the quote below sums up what carries over: instructions, connected apps and knowledge files. Migration uses the latest published version, so unpublished drafts do not transfer.

Five things stay behind or change:

• Custom actions. "GPT custom actions do not transfer through the migration workflow." Rebuilding one may need an existing app or a custom MCP server.
• The model. "The GPT's selected model does not carry over. In Enterprise workspaces, Enterprise defaults apply."
• Access. "An individual migration starts with a private plugin. Existing GPT users do not gain access automatically."
• Behavior. "A migrated plugin may respond differently." Test before you share.
• Conversations. Past chats stay in ChatGPT and remain accessible. They do not move into the plugin.

The original GPT stays visible for now: it "becomes read-only after migration and its creator cannot delete it." On the retirement date it becomes inaccessible like every other GPT.

Treat a migrated plugin as a new tool with a familiar name: it needs testing, sharing and an owner.

4. Inventory Your GPTs This Week

Start with a complete list. In an Enterprise workspace, "Retrieving the full workspace GPT inventory currently requires a workspace Owner," per the FAQ, so go to your Owner first. Everywhere else, ask each person to open My GPTs and send you what is there, including anything on a personal account.

Record seven things for each GPT:

1. Owner. Who built it, and who maintains it now.
2. Users. Who used it in the last 90 days, and for what task.
3. Knowledge files. Which documents it holds, and where the master copies live.
4. Actions and apps. What it connects to or calls, especially anything that reads or writes deal, borrower or investor data.
5. Data sensitivity. Whether it sees MNPI, LP or family information, borrower files or personal data.
6. Records. Whether your retention policy treats its outputs as records.
7. Where it was built. The firm's Enterprise or Business workspace, or someone's personal account.

The last item matters most. A GPT on a personal account holds firm documents on consumer terms. On ChatGPT Free, Plus and Pro, "data sharing is enabled by default" unless the user turns off Improve the model for everyone (OpenAI data controls FAQ). Those GPTs are shadow AI with a deadline attached.

Then sort the list into four piles: migrate, move, rebuild and retire. A GPT with no owner and no users in 90 days goes straight to retire.

5. Option One: Migrate to a ChatGPT Plugin

If your firm stays on ChatGPT, migrating in place is the shortest path. The FAQ gives the step: "go to My GPTs and select Migrate to plugin for the GPT you want to migrate." In Enterprise workspaces, "Bulk migration is available through the ChatGPT Admin plugin."

A plugin holds more than a GPT did. OpenAI describes plugins as "bringing together the tools, knowledge, and reusable workflows you need to get work done" (plugins in ChatGPT). The skill your GPT's instructions become is one of the "reusable, shareable workflows that help ChatGPT complete specific tasks more consistently" (skills in ChatGPT).

The tool does the copying. Four steps are still yours:

1. Test it. Run three real tasks through the plugin and compare the output with the old GPT's, because a migrated plugin may respond differently.
2. Rebuild the actions, with an existing app or a custom MCP server. In Enterprise, "Use plugins" is on by default while "Upload plugins" and "Create plugins with MCPs" are off (plugins in ChatGPT), so an admin decides who can build that way.
3. Share it. "Users on Business and Enterprise workspaces can share plugins with other members of their workspace," per the same plugins article. The GPT's old users get nothing until you do.
4. Check the model. In Enterprise workspaces the plugin runs on Enterprise defaults, whatever model the GPT used.

The business terms carry on. OpenAI says: "By default, we do not use your business data for training our models" (OpenAI enterprise privacy). ChatGPT Team was renamed ChatGPT Business on August 29, 2025 (OpenAI Help Center), and a standard Business seat costs "$20/month if billed annually. $25/month if billed monthly" (OpenAI pricing). Enterprise is priced by quote. If your inventory found GPTs on personal accounts, a business workspace is their first stop.

More on the platform for deal teams is in ChatGPT for private equity. The migration is the quick part. The testing is the work.

6. Option Two: Move It to Claude

If your firm already uses Claude, or has chosen it, the retirement is a natural moment to consolidate. A GPT's parts map onto Claude's closely, and your team copies them across:

• A project, for a GPT that is mostly reference material plus instructions, such as a DDQ answer library. "Projects allow you to create self-contained workspaces with their own chat histories and knowledge bases" (Anthropic Help Center). The knowledge files go into the project.
• A skill, for a GPT that does one repeatable task, such as an IC memo section in house format. Skills "are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks" (Anthropic Help Center).
• A plugin, for a GPT that also called other systems. "Each plugin bundles skills, connectors, and sub-agents into a single package" (Anthropic Help Center), and connectors do the job the GPT's apps and actions did.

Sharing runs through the organization on the business plans. "Organization-wide skill management is available to Team and Enterprise plans" (Anthropic Help Center), and "Plugin marketplaces let Team and Enterprise plan owners distribute curated plugins to everyone in their organization" (Anthropic Help Center).

Keep firm data on those plans. Anthropic's Privacy Center says that by default it will not use inputs or outputs from its commercial products, Claude for Work included, to train its models, though conversations sent as thumbs-up or thumbs-down feedback may be used. Its pricing table shows model training as "Opt-out" for Free, Pro and Max and "None by default" for Team and Enterprise.

The guide to Claude skills for private equity shows how a firm builds and shares them. Moving takes more effort than migrating in place, and it pays off when the firm's work is already heading to Claude.

7. Option Three: A Microsoft 365 Copilot Agent

If the firm lives in Microsoft 365 and the GPT's documents already sit in SharePoint, a Copilot agent may be the better home. Microsoft's description: "Agents extend the functionality of Copilot by acting as specialized AI assistants tailored to specific domains" (Microsoft Learn).

Two limits shape the move:

• Actions. "Agent Builder doesn't support authoring actions that integrate external services" (Microsoft Learn), so a GPT with custom actions needs Copilot Studio.
• Licensing. In a tenant without Copilot licenses, "Usage billing applies to agents grounded in tenant data, while agents grounded in public data or instructions are free to use" (Microsoft Learn). A Copilot license gives full agent access.

The second limit matters to deal teams: a GPT that answers from the firm's own documents becomes an agent grounded in tenant data, so price its usage before you move a busy one.

The Copilot guide for private equity covers the rest of the platform. Copilot fits best when the documents are already there, because moving them to SharePoint for one agent is a bigger project than the agent.

8. Option Four: Rebuild It Properly, or Retire It

Custom actions have to be rebuilt wherever a GPT goes, which makes the retirement a good moment to ask whether they belonged in a GPT at all. Before you rebuild one that writes to a CRM, a data room or a loan administration system, answer four questions: who can trigger it, what it can change, what gets logged, and who checks the result.

If the answers call for access controls, structured outputs and an audit trail, a custom build is the right destination. The guide to Claude projects vs a custom build sets out where off-the-shelf tools stop.

Then retire the rest on purpose. Retiring costs nothing and removes a tool nobody supervises. Before you let one go:

• Save its instructions in a document, with the owner and the date.
• Confirm every knowledge file has a master copy in the firm's own document system.

A simple test helps: if nobody used it in 90 days and nobody objects when you announce its retirement, let it go. Every GPT you retire is one less thing to test, share and supervise.

9. The Five Destinations Side by Side

Put the options next to each other and most GPTs sort themselves. The platform your firm already pays for usually settles it.

If most of your GPTs point somewhere other than ChatGPT, you are facing a platform decision. The comparison of ChatGPT, Copilot and Claude for private equity covers the trade-offs.

Decide per GPT, then look at the pattern. The pattern is the real decision.

10. Records, Knowledge Files and Conversation History

Three kinds of material sit inside a GPT program, and they part ways on the retirement date:

• Conversations. "Your existing conversations with custom GPTs will remain accessible after retirement," says the FAQ. They stay in ChatGPT, kept under your workspace's retention settings, and they do not move into a migrated plugin.
• Knowledge files. Migration copies them into the plugin's reference files. If you don't migrate a GPT, assume its files go with it, and keep master copies in the firm's own systems.
• Instructions. Save each GPT's instructions with its owner and the date. They are the specification for whatever replaces it, and the record of what the tool was told to do.

Then ask your CCO whether the firm treats AI conversations or outputs, such as LP letter drafts or credit memo sections, as records, and where they should be kept. The CCO and counsel decide.

Your past chats survive the retirement. An unmigrated GPT's setup does not, so copy it now.

11. A Six-Week Plan to December 11

Six weeks, counted back from December 11:

1. Week 1, count. Get the full list from your workspace Owner and each person's My GPTs, and fill in the seven items.
2. Week 2, decide. Put every GPT in a pile: migrate, move, rebuild or retire. Confirm your date in the workspace notices.
3. Week 3, save. Export instructions, confirm master copies of knowledge files, and agree retention with the CCO.
4. Week 4, build. Migrate the keepers to plugins, or set them up in Claude or Copilot. Rebuild any actions.
5. Week 5, test and share. Run each replacement on real tasks beside the old output, then share it with the people who used the original.
6. Week 6, switch over. Tell users where each tool now lives, announce the retirements, and book a first-quarter review.

Firms with a deferral to February 11, 2027 can run the same plan more slowly. New GPT creation has already ended in affected Enterprise workspaces, so new work should start in plugins or on the firm's chosen platform.

Six weeks is enough for a firm with a dozen GPTs. A firm that starts in December will be choosing what to lose.

12. Where to Start

Today, check the retirement date in your workspace notices and ask your ChatGPT workspace Owner for the full GPT list. Then:

1. Send one email asking everyone who built a GPT, on any account, to reply with its name and purpose.
2. Fill in the inventory, including where each GPT was built.
3. Retire the GPTs nobody uses, after saving their instructions and files.
4. Pick a destination for the rest from the table in section 9.
5. Book the testing time before anyone presses Migrate to plugin.

If the pattern points to a platform change, make that decision once, for the whole firm, before anything moves.