Who Owns Your Company's AI? Why the System Should Live in Your Accounts

Your company owns its AI only if the accounts, the knowledge, the written methods and the connections sit in its own name. If they sit with a supplier or an employee's personal subscription, you are renting them. The test is simple: if that supplier or person left tomorrow, would the system keep working?

Who owns your company's AI?

Whoever controls the accounts, files and connections it runs on. Paying for a tool is not the same as owning the system around it. If the business knowledge, the instructions and the logins live in a supplier's platform or a staff member's personal account, the business depends on them, however much it pays.

Most businesses never ask the question. AI arrived through individuals, then through suppliers, and each made sensible choices for themselves. A salesperson built a great prompt library in their own account. An agency set up workflows in its own platform. Both work well, until someone leaves.

Ownership is not about doing everything in-house. A partner can build and run the system. The question is only where the system lives while they do.

Where does AI lock-in actually happen?

Lock-in happens in five layers. The account holds the data and history. Knowledge is what the AI knows about your business. Methods are the written instructions for each job. Connections link AI to your systems, and the model is the AI itself. Each layer can sit in your name or someone else's.

The model layer worries people most and matters least. Models improve and change prices constantly, so switching between them should be routine. Knowledge and methods are the expensive layers to lose, because they took the longest to build.

Why are personal AI accounts a form of lock-in?

Because whatever an employee builds in a personal account belongs, in practice, to them. Their prompts, saved context and conversation history leave when they do. The business also has less control over how that data is handled, since personal plans often run on consumer terms rather than business terms.

The gap is wide. MIT NANDA's research found workers from over 90% of companies reported regular use of personal AI tools for work. Only 40% of companies said they had bought an official subscription.

Terms differ too. When Anthropic updated its consumer terms in 2025, it said the changes "do not apply to services under our Commercial Terms", which cover Claude for Work. Consumer users choose whether their chats can be used to improve Claude. A business plan gives the company the controls instead.

Knowledge loss is the bigger cost. According to Panopto's 2018 survey, 42% of institutional knowledge is unique to the individual holding it. The guide to offboarding in the AI era covers what to do when someone leaves.

How do open standards reduce AI vendor lock-in?

Open standards let the same methods and connections work across different AI tools, so changing tools does not mean rebuilding. The Model Context Protocol, known as MCP, standardises how AI connects to other systems. Agent Skills standardises how written methods are packaged. Both are now supported well beyond the companies that created them.

MCP started at Anthropic and is now governed more widely. According to Anthropic, it donated MCP to the Agentic AI Foundation, a fund under the Linux Foundation, in December 2025. It said ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code had adopted it. The MCP explainer covers how it works.

Skills follow the same path. The Agent Skills specification describes "a lightweight, open format for extending AI agent capabilities", with the promise to "build a skill once and use it across any skills-compatible agent". A method written this way is not tied to one vendor.

Standards help only if you use them. A method written in an open format but stored in a supplier's account is still locked in.

What should a contract with an AI supplier say?

It should say that accounts are in your name, that knowledge and methods are stored in files you own, that connections use access you grant and can revoke, that everything is handed over in a usable format, and that the system keeps working without the supplier. Ask for all five in writing.

1. Accounts. AI subscriptions and platforms are registered to the business, with the business as administrator.
2. Files. Knowledge and methods live in your storage, not only in the supplier's tools.
3. Access. Connections use credentials you control, starting read only.
4. Handover. On exit, everything is handed over in a format another person or AI can use.
5. Continuity. Named workflows keep running if the relationship ends.

Official guidance supports this. According to Australia's National AI Centre, organisations should make it clear "who is responsible for each part of the AI supply chain". The guidance also advises keeping critical functions able to continue "even if your AI systems malfunction or are being retired". A supplier exit is one version of that.

How do you test whether you own your AI?

Run the leaving test. Imagine your main AI supplier, and separately your most AI-savvy employee, left tomorrow. List what would stop working, what knowledge would go with them, and which logins you could not reach. Anything on that list is rented, not owned, and is the place to start fixing.

Five questions make the test concrete:

• Accounts. Who is the administrator of every AI tool we pay for?
• Knowledge. Could a new AI tool read our business facts tomorrow?
• Methods. Are our best prompts and workflows written down where we can reach them?
• Access. Could we revoke every AI connection to our systems today?
• Continuity. Which workflows would still run next week?

The guide to outputs versus ownership covers the related choice between buying finished work and owning the system.

Skein builds each client's Company OS in files and accounts the business owns, so it keeps working if Skein leaves, and teams keep Claude, ChatGPT or whichever AI they prefer; the Company OS page shows the three layers. That suits a growing B2B business that wants help building its AI but intends to keep it. A business that prefers a fully managed service, with the supplier holding everything, may find a conventional agency simpler.

Frequently asked questions

What is AI vendor lock-in?

AI vendor lock-in is when a business depends on one supplier or tool so heavily that leaving would mean rebuilding. It usually happens because business knowledge, written methods, connections or accounts sit in the supplier's platform rather than the business's own files and accounts.

How do you avoid AI vendor lock-in?

Keep AI accounts in the company's name, store knowledge and methods in files you own, grant and revoke connections yourself, prefer open standards such as MCP and Agent Skills, and require a usable handover in every supplier contract. Then test regularly what would keep working if a supplier left.

Who owns what an employee builds in a personal AI account?

In practice, the employee keeps it. Prompts, saved context and history in a personal account leave when they do, and the business has little control over how that data is handled. Moving staff onto business plans the company administers keeps that work inside the business.

Does using ChatGPT or Claude create lock-in?

The model itself is the easiest layer to switch, as long as your knowledge and methods sit in your own files rather than inside one product. Lock-in grows when everything is built around one tool's proprietary features. A shared layer that several AI tools can reach keeps your options open.

What should happen to our AI system if we change suppliers?

It should keep working. Accounts should already be in your name, knowledge and methods in your own files, and connections under your control, so a new supplier or an internal person can take over. Write this into the contract before work starts, not at the point of leaving.

Related reading