A summary for funders
PL R&D's Sovereign AI program helps governments run AI on their own terms. In each country we take one public problem to a working pilot on an open model the government controls, and train the local team that will run it. We aim for three to five countries over 24 months. Funding pays for the pilots, their independent evaluation and the lessons the next country can reuse, so a government sees AI work for its priorities before it commits its own budget.
Three reasons, with the evidence behind each
Capacity is the help they want first
Ministries want to choose, run and check AI themselves. Most offers they get are a product or a plan. We build the skills on a real public task.
In more than 1,500 written submissions to the UN's first Global Dialogue on AI Governance, governments were the only group to rank capacity-building first. Most others put safety first.
Open models already do real public work
A government can now host and adapt an open model, such as Switzerland's Apertus, for work like translation and drafting. What most countries lack is the team to run one safely.
The Canton of Ticino translates sensitive government documents on Switzerland's open model Apertus, without relying on commercial providers.
Rented AI answers to someone else
Whoever runs a government's AI shapes what its officials see. Renting all of it hands over the data, the audit trail and the choice of model. Capacity built now keeps them at home.
said Microsoft France's director of public and legal affairs under oath, asked to guarantee French data would never reach US authorities without French consent. He added it had never happened yet.
In each country, around the government's own team
An open model hosted where the government chooses, adapted to its languages and data, and engineers to build the service with its team.
Academic partners test models on the country's priorities and languages, then measure the pilot against the current process. We publish the results, good or bad.
Training and side-by-side work for the officials and engineers who will run and improve the system after the pilot ends.
A program manager, gatherings where pilot teams compare notes, and an open playbook the next country can start from.
In each country, from a named problem to a decision
Months 0 to 3. One public service or decision to improve, its owner, and a measure of success.
Months 3 to 6. Data rules, oversight and red lines first, then the model and where it runs.
Months 6 to 12. A live service used by over 1,000 people for at least 30 days, tested against the current process.
Scale, change or stop, on the evidence. The code, the know-how and the team stay with the government.
What every funder gets: a measure of success agreed before work starts, results reported against it, good or bad, and open code and lessons that any country can reuse.
Some pilots will end at the decision step, and we will publish why. One pilot does not make a country self-sufficient in AI. That takes people who stay, a budget line and years of use. The program pays for the first working system and the evidence a government needs to fund the rest.
Tell us what you want your funding to change, and we will show you where it fits in the program. research@protocol.ai
David Dao and James Tunningley, co-leads, Economies & Governance.
Also drawn on: PL R&D, Why sovereign AI, October 2026 · PL R&D, Sovereign AI, government pilot partnership, October 2026 · How PL R&D Accelerates Breakthroughs, 14 July 2026 · Programmable Economies & Governance, 23 July 2026
Protocol Labs R&D · Economies & Governance focus area. Programmable infrastructure for more efficient, equitable coordination at global scale. Evidence checked 22 October 2026.