TL;DR

OpenAI has published its approach to localising AI systems for different countries. The “OpenAI for Countries” programme adapts ChatGPT to local languages, laws, and cultural norms while maintaining red-line safety principles that cannot be overridden by localisation.

Sovereign AI, Not Just Translation

OpenAI says governments consistently tell the company they want “sovereign AI they can build with us, not just systems translated into their language.” Most countries cannot develop frontier AI models from scratch, so the challenge is adapting the best available systems for local contexts.

The company is already piloting a localised version of ChatGPT for students in Estonia, incorporating local curricula and teaching approaches as part of its ChatGPT Edu work. Further pilot localisation efforts with other countries are under way.

Red Lines That Cannot Move

OpenAI’s Model Spec — a public document defining how its models should behave — includes “red-line principles” that apply to all deployments, including localised versions. These prohibit models from enabling violence, weapons of mass destruction, terrorism, persecution, mass surveillance, or undermining of democratic processes.

For consumer-facing products like ChatGPT, OpenAI adds further commitments: localisation may affect language or tone, but cannot change the substance or balance of facts presented. Any content removed due to legal requirements must be transparently flagged to the user, specifying what was removed and why — without revealing the redacted content itself.

What This Means in Practice

The approach sits between two extremes. On one side: a single global model that ignores local context. On the other: fully independent national AI systems that few countries can afford to build. OpenAI’s middle path offers adapted versions of its frontier model that respect local needs while keeping safety guardrails intact.

For governments weighing AI strategy, the proposal raises familiar questions about sovereignty, dependency, and where to draw the line between customisation and control.