# What is the difference between sovereign AI, public AI and personal AI?

Date: 2026-10-29 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/sovereign-public-personal-ai/

**Answer:** Sovereign AI is about who controls AI at the level of a country or an organization, public AI is AI built by public institutions for the public interest, and personal AI is AI that runs on your own device or works only for you. The three answer the same question at different scales: who holds the keys to the AI you depend on, and to your data.

Three phrases now come up whenever AI meets politics or privacy: sovereign AI, public AI and personal AI. They sound like slogans, but each one names a real choice about who builds the AI, where it runs and who can see what you type into it. Once you can tell them apart, a lot of the news about AI investments, national strategies and "run it on your own computer" guides falls into place.

The simplest way to hold them together is to think of three scales: a country, a society and a person. At each scale the question is the same. Who controls the model, the computers it runs on and the data that passes through it?

## Sovereign AI: control at the level of a country

Sovereign AI means a country, or a large organization, being able to build, run and govern the AI it depends on without relying entirely on foreign companies. Canada's national AI strategy, *AI for All*, describes AI as a chain of layers, from electricity and chips up to data centres, models and applications, and points out that much of that chain sits outside the country. Its answer is to build some layers at home, partner with trusted allies for others and buy the rest.

That is the logic behind Ottawa's plan to put up to $1.6 billion into Cohere, announced on October 29, 2026: a Canadian company that trains its own models, commits to stay headquartered in Canada and adds computing capacity here. Sovereign AI is not about banning foreign tools. It is about having a credible alternative for the work where control matters most, such as health records, defence or government services.

## Public AI: built by public institutions, for the public

Public AI goes one step further: the model itself is treated like public infrastructure, the way roads or libraries are. The clearest example so far is Apertus, released on September 2, 2025 by two Swiss universities, EPFL and ETH Zurich, with the Swiss National Supercomputing Centre. Everything about it is published: the model weights, the training data recipes, the code and the documentation, under a licence that allows commercial use. It was trained in more than 1,000 languages, including Swiss German and Romansh, which commercial labs tend to overlook.

A public model can be inspected by anyone, adapted for a local language or a public service, and run by a hospital or a school without sending data to a private company. The trade-off is that public projects usually have smaller budgets than the largest commercial labs, so the models are not always the most capable on the market.

## Personal AI: AI that works only for you

Personal AI is the scale closest to you. It covers two things that often go together. The first is running a model on your own computer or phone, so your prompts never leave the device; tools such as Ollama and LM Studio made that practical for ordinary laptops, and our guide to [where your AI should run](/en/learn/cloud-local-where-ai-run-2026/) compares the options. The second is a personal agent: an assistant that knows your files, calendar and accounts and acts on your behalf.

Personal AI gives you the most control over your data, but also the most responsibility. A local model needs a capable machine and some setup. An agent with access to your email and bank accounts can do real damage if it is tricked or misconfigured, which is why [the risks of a personal agent](/en/learn/go-wrong-personal-agent-that-has-accounts/) deserve a careful read before you install one.

## The three side by side

| | Sovereign AI | Public AI | Personal AI |
|---|---|---|---|
| Scale | A country or a large organization | A society | One person or a small team |
| Who controls it | A government or domestic companies | Universities and public institutions | You |
| Example | Canada's stake in Cohere | Switzerland's Apertus | A model running on your laptop |
| Main benefit | Data stays under local law | Anyone can inspect and reuse it | Your data never leaves your device |
| Main limit | Costly, may lag the leaders | Smaller budgets | Needs hardware, setup and care |

The categories overlap. Apertus is public, and it also serves Swiss sovereignty; downloaded onto your own computer, it becomes personal AI as well. An open-weight model is the bridge between all three, because whoever holds the weights can run it where they choose. Our page on [open-weight and closed models](/en/learn/open-weights-vs-closed-licences/) explains the licences.

> **Try it:** List the three AI tools you use most. For each, write down which scale it belongs to (a foreign company, a public project or your own device) and one task you would not give it because of that.

## Which kind of AI should you choose for your own work?

Start from the data, not the tool. For everyday tasks with nothing sensitive in them, a mainstream hosted assistant is usually the most capable and the least work. When the data is covered by a law or a contract, such as health information, client files or anything under Quebec's Law 25, look for a provider that processes data in Canada or under the rules you are bound by. That is where sovereign options matter.

When the data must not leave your hands at all, run an open-weight model locally and accept a smaller, slower model in exchange for full control. And if you want to support or study AI that anyone can inspect, try a public model such as Apertus. Many people end up using all three, each for the work it fits.

## Sources

- Canada's National Artificial Intelligence Strategy: AI for All: https://ised-isde.canada.ca/site/ised/en/node/1046
- Apertus: a fully open, transparent, multilingual language model: https://ai.epfl.ch/apertus-a-fully-open-transparent-multilingual-language-model/
- Canada announces landmark investment in Cohere to scale secure, sovereign AI: https://www.newswire.ca/news-releases/canada-announces-landmark-investment-in-cohere-to-scale-secure-sovereign-ai-814597810.html

---

From the Guide to Artificial Intelligence, by Robert Scales, published by Holistic Hybrid Inc. Share and adapt it under CC BY-NC-SA 4.0 (https://creativecommons.org/licenses/by-nc-sa/4.0/): credit the guide, link to https://guidetoartificialintelligence.com/en/learn/sovereign-public-personal-ai/, no commercial use without permission, and keep the same licence.