System One AI & THEMIS

System One models are designed to make fast, structured decisions inside software rather than generate open-ended text for people: they answer typed questions with a choice, a score or a yes/no probability that code can act on directly. Jev, introduced by TypeSafe AI in September 2026, is the first public model positioned as one, and the category is relevant to enterprise workflows such as classification, routing, risk scoring, security checks, sensitive-data detection and policy enforcement.

A typed output still needs governing. The EU AI Act creates no special category for System One models, so obligations depend on each organisation's role and what the AI system is used for, and using a US-hosted model lawfully in Europe is not the same as deploying a sovereign one. These articles start with what Jev is, then where System One AI fits and what it takes to govern it, and end with THEMIS, the sovereign System One model Colchix is building for real-time data protection and AI governance in Europe.