Data sovereignty: why you can't depend on one AI platform

AI tools are becoming part of everyday work. That is useful, but it creates a quiet risk: the useful context can end up scattered across chat histories, custom assistants and automation builders that you do not control.

Data sovereignty is not about distrusting every provider or refusing cloud tools. It is about deciding what must remain portable: the knowledge, rules and decisions that make your business work.

What business context should you control?

Keep the durable source material in formats your team can open without a particular AI subscription:

• product and pricing rules;
• customer-facing templates and approved examples;
• operating procedures and validation checks;
• project decisions, meeting notes and lessons learned;
• a short brief explaining how your company works and what an assistant may not do.

Markdown is often a good default because it is readable, versionable and easy to move. A spreadsheet, PDF or exported CRM record can also be appropriate. The important point is that the original remains accessible outside the AI interface.

The practical model: one source, many tools

Avoid copying the same company profile into five separate custom bots. Instead, keep a canonical folder or second brain, then give each tool the small, relevant slice it needs.

For example, a sales assistant might receive the current price list, an approved proposal and a short SOP for preparing a quote. A content assistant receives the editorial brief and relevant source notes. Neither needs an unbounded export of your company.

This model has three benefits:

1. You can change tools without rewriting your business. If a provider changes a feature or no longer fits, your working context remains usable.
2. You improve quality. A clear source file is easier to review than a hidden instruction buried in an old conversation.
3. You reduce risk. Sending only what a task needs limits unnecessary exposure of confidential information.

A small-business migration checklist

Start with one workflow, not an enterprise-wide knowledge project.

1. List the files and decisions the workflow relies on.
2. Move the durable material into a named, human-readable folder.
3. Write a short SOP: inputs, steps, expected output, validation and when to stop.
4. Export or save any important instructions currently trapped in a chat or automation tool.
5. Test the same workflow with a second approved tool using those files.

If the second tool can produce a useful draft after a short briefing, you have reduced dependency without adding much administration.

Portability is not the same as data safety

Portable files still need sensible access controls, backups and retention rules. Before sending personal, customer or commercially sensitive data to an AI service, check the provider’s current terms, choose the appropriate account settings, and involve a privacy or legal adviser where the use case requires it. Do not promise a client that a tool is compliant simply because it is popular or based in a particular region.

The objective is simple: your AI tools should help your business think and execute, while the business knowledge stays understandable, reviewable and yours.

Next steps

Start by building a second brain in Obsidian for AI, then use AGENTS.md as your AI brief to make the operating rules explicit.