Skip to content

Private AI8 min read

Private AI vs ChatGPT Enterprise: who owns your AI knowledge?

Private AI vs ChatGPT Enterprise compared: data ownership, AI vendor lock-in, model retirements, switching costs and EU, US and Canadian rules, told through the rabbit Walt Disney created but did not own.

Manuj AggarwalFounder & CIO, TetraNoodle
Illustration: a stack of knowledge layers with a key above it connects to a router that switches between three AI model chips, with an open padlock showing there is no lock-in.

Most business owners now use AI in some form. After hundreds of conversations with leaders, one concern keeps coming up: who truly owns this intelligence?

It matters because, as I said in a recent live session, “the true wealth of any organization, any human, any individual is our data. Even your money is basically a data item in your bank’s database.” Your wealth is your data, and the engine that generates it is intelligence: yours, your team’s, and now the machine’s.

The rabbit Walt Disney created but did not own

In 1927, before Mickey Mouse, Walt Disney created a character called Oswald the Lucky Rabbit. His small studio made 26 Oswald cartoons, distributed through Universal. Audiences loved them.

In early 1928, Walt travelled to New York to renew the contract and ask for more money to make more films. He was refused. Then he found out that, under the contract, he did not own the rights to Oswald. Universal did. He was the creator, but not the owner.

He did not give up. On the trip home he came up with a new character: Mickey Mouse. The first Mickey shorts did not take off. Then, in November 1928, Steamboat Willie opened in New York as one of the first cartoons with synchronised sound, and Mickey became a household name.

Oswald stayed with someone else for almost 80 years. In 2006, Disney finally got him back from NBC Universal, as part of an unrelated deal that let sportscaster Al Michaels move from ABC and ESPN to NBC. Disney wanted its own creation back badly enough to trade a star broadcaster for a cartoon rabbit.

Intellectual property, ideas, intelligence is the future currency. If you don’t know who owns that currency, then you may be building somebody else’s empire.

Manuj Aggarwal, live session

Who owns your data in ChatGPT Enterprise?

Here the news is better than many people assume. OpenAI’s Services Agreement covers ChatGPT Business, Enterprise and the API. Under it, the customer retains ownership of its inputs and owns the outputs. OpenAI does not use customer content to improve its services unless the customer agrees, and it deletes customer content within 30 days after the contract ends unless the law requires otherwise.

So on paper, you own your data. But owning the data is not the same as owning the system you built around it. Over months of use, your team also builds:

  • the prompts and instructions that encode how your business works,
  • custom GPTs and assistants tuned to your processes,
  • agents connected to your drive, CRM and inbox,
  • and the habits of a team that has learned one tool’s quirks.

That is your AI knowledge, and almost none of it moves cleanly to another vendor. OpenAI’s own help centre says self-service export is not available in ChatGPT Business workspaces, and that cancelling deactivates the workspace. Owners can export a member’s data only “when the Export data option is available.”

What AI vendor lock-in really looks like

If you have ever tried to change your accounting software or CRM, you know switching is hard. With AI, as I said in the session, “it’s not just a software provider of these applications, it’s actual intelligence.”

Now imagine your provider changes its terms tomorrow. Or raises prices tenfold. Or retires the model your workflows were tuned for. Could you leave? Could you take your knowledge with you? Could you be sure it was deleted?

These are not hypothetical events. In February 2026, OpenAI retired GPT-4o, GPT-4.1 and o4-mini from ChatGPT. Its API deprecation page lists a steady run of model shutdowns, with as little as about two weeks’ notice for preview models.

of executives say switching their primary AI vendor or model would be difficult
71%
do not fully understand their AI dependencies
91%
say a 7-day vendor outage would cause severe or critical disruption
81%

Those figures come from IBM’s 2026 study of 1,000 senior executives, The Calculus of AI Sovereignty. Respondents named price increases, usage restrictions, model deprecations and performance drops among the disruptions they had already faced.

And it is not only people using AI anymore. Agents and bots now take actions on your behalf. Picture your business in 6, 12 or 24 months, more dependent on these platforms every week. How are you protecting that dependency?

Locked in versus a private AI layerTop: documents, email and CRM data all flow into one AI vendor, so the business depends on that vendor's terms, prices and models. Bottom: the same data stays in a private AI layer the business controls, and a router sends each request to whichever model fits, so models can be swapped.1 · Locked inDocsEmailCRMOne AIvendortheir termstheir prices2 · A private layer you ownDocsEmailCRMYour privateAI layeryour rulesRouterModel AModel BOpen model
Same data, two architectures. In the second one, the AI model is a part you can replace.

Private AI vs ChatGPT Enterprise, side by side

ChatGPT EnterprisePrivate AI
Where your knowledge livesOn OpenAI’s platformIn an environment you control: your office, your cloud account, or a hosted space under your control
AI modelsOpenAI’s modelsAny model, routed per task, including low-cost and open models
Training on your dataNot by defaultOnly if you decide to, on your own models
How data gets inPeople paste or upload, or use connectorsConnects to your systems and fetches just enough data for each question
Switching to another vendor’s modelsNot possibleSwap a model without rebuilding workflows
LeavingLimited export, workspace deactivatedYour data and logic stay with you
Cost controlPer-seat licencesRoutes simple tasks to cheaper models, pays for power only when needed
Best forFast, company-wide chat productivityCore know-how, regulated data, multi-model strategy

When ChatGPT Enterprise is the right choice

Be honest about this. If you mainly need secure, company-wide chat for writing, research and analysis, and you are comfortable building on one vendor, Enterprise is a strong, well-documented option. It also beats letting staff use personal accounts by a long way.

When a private AI is the better fit

  • Your edge is how you work: your pricing logic, your client history, your decades of judgment.
  • You handle regulated or sensitive data: health, finance, legal, education, government.
  • You operate in the EU or Canada and want clear control over where data is processed.
  • You want to use the best model for each job, not one model for every job.
  • You need to be able to leave any vendor, including us, without losing what you built.

The law is starting to back portability

Regulators are moving toward making exits possible.

  • European Union. Since 12 September 2025, the EU Data Act has required cloud and data processing providers to support switching. From 12 January 2027, switching charges, including data egress fees, must be removed entirely. Under GDPR, you also need clear processor terms for any personal data an AI vendor handles.
  • United States. The FTC has warned that quietly changing terms to start using customer data for AI training may be unfair or deceptive. Read your vendors’ change notices, and keep copies of the terms you signed.
  • Canada. PIPEDA’s safeguarding principle, and Quebec’s Law 25 assessment before personal information leaves the province, both reward architectures where you know exactly where data goes.

Rules help, but they move slowly. Architecture is the protection you control today.

Two real conversations about ownership

A team of eight to ten people. They used Claude, Gemini, Google Docs and Notion, getting a little out of each. Their frustration was that it took more time to work out where to put their data than to solve the actual problem. As they said, nobody wants to use all these tools. They want to solve their problem, securely and privately, and stay independent of the technology underneath.

They then asked us a fair question: what if they want to leave us? My answer was that in the AI age, trust and credibility are the most important assets. We do not lock anyone in. If a client finds something better, they can export their data or ask us to delete it. The door stays open.

A family office and venture investor. Their concern was different: how to pass on decades of judgment to portfolio companies and the next generation. How they evaluate risk, how they communicate, how they decide. Those are cultural, human things. In a private AI, that wisdom can be captured once and shared safely, so that even an intern can draw on 50 years of experience in a way that makes sense to them. That is exactly the kind of knowledge you never want sitting in someone else’s empire.

Seven questions to ask any AI vendor before you commit

  1. Who owns our inputs, outputs, prompts, custom assistants and fine-tuned models? Show me the clause.
  2. Can we export everything, in a usable format, without your help? Let us test it now.
  3. Can we delete everything, and get confirmation in writing?
  4. Which models do we depend on, and how much notice do you give before retiring one?
  5. Where is our data stored and processed, and does a copy ever leave our chosen region?
  6. Who can access our data, including your staff, subcontractors and connected agents?
  7. What happens to cost if usage triples or your prices change?

If a vendor cannot answer these clearly, you have learned something important before signing.

Build your own empire, not someone else’s

Walt Disney learned the hard way that creating something is not the same as owning it. Your team is creating valuable intelligence every day it uses AI. Make sure it accumulates somewhere you control.

We build private, model-independent AI environments that connect to the systems you already use, with permissions and human approval designed in from day one. If you want to talk through what that would look like for your business, tell us your goal, or read how we work.

Sources

Questions people ask

What is the difference between private AI and ChatGPT Enterprise?

ChatGPT Enterprise is a managed workspace on OpenAI's platform: your data is not used for training by default and you get admin controls, but it runs on OpenAI's models, terms and infrastructure. A private AI is an environment you control. It connects to your own systems, keeps your data under your rules, and can route each task to whichever AI model fits, so you can change models without starting over.

Who owns the data I put into ChatGPT Enterprise?

Under OpenAI's Services Agreement, business customers retain ownership of their inputs and own the outputs, and OpenAI does not use customer content to improve its services unless the customer agrees. Owning the data is not the same as owning the system around it, though: the prompts, custom GPTs, agents and workflows your team builds live on OpenAI's platform.

What is AI vendor lock-in?

AI vendor lock-in is when your processes, data and know-how become so tied to one AI provider that switching becomes slow, risky or expensive. In a 2026 IBM study of 1,000 senior executives, 71% said switching their primary AI vendor or model would be difficult.

How do I avoid AI vendor lock-in?

Keep your data in systems you control, connect AI to it rather than copying it into a chat tool, use a model-independent layer that can route to more than one AI model, get export and deletion terms in writing, and test that you can actually export your data before you depend on a tool.

Can I export my data from ChatGPT Business?

Only in a limited way. OpenAI's help centre says self-service export is not available in ChatGPT Business workspaces, and owners can export a member's data only when the Export data option is available. Cancelling deactivates the workspace rather than handing you a complete archive, so plan your exit before you need it.

Does the EU Data Act help with switching AI or cloud providers?

Yes, for cloud and data processing services. Since 12 September 2025 the EU Data Act requires providers to make switching possible, and from 12 January 2027 switching charges, including data egress fees, must be removed entirely.

Keep reading

All posts

Want this working in your business?

Tell us what you want to get done. We will tell you if AI can help.