28/10/2026
How do you check if an AI avatar is really sovereign?
Ask the vendor which level it would claim on the European Commission’s own 0–4 scale.
There’s still no EU certificate that says an AI system is sovereign. But there is now a published scale, and buyers can point to it.
• 𝗦𝗘𝗔𝗟-𝟬 𝘁𝗼 𝗦𝗘𝗔𝗟-𝟰: the Commission’s scale. In April 2026 it used SEAL-2 as the minimum to screen bidders for its sovereign cloud contracts.
• 𝗕𝗦𝗜 𝗖𝟯𝗔: Germany’s federal cyber security office turned that scale into voluntary, testable criteria, including a yearly disconnection test.
• 𝗧𝗵𝗲 𝗖𝗹𝗼𝘂𝗱 𝗮𝗻𝗱 𝗔𝗜 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗺𝗲𝗻𝘁 𝗔𝗰𝘁: a proposal that would put four assurance levels into law. Still a proposal.
For an AI avatar, these criteria grade the infrastructure it runs on. So ask any vendor, us included, for a level and a criterion number, not an adjective. Which level would you claim, and what stops you claiming the next one up?
Part 2 of our series on sovereign AI avatars. Link in the first comment 👇