The quiet regionalisation of AI governance

AI risks are not uniform but vary sharply by region. Global frameworks are often too broad to meaningfully address specific risks. Regional governance allows blocs to target the issues that are most acute in their own context. Southeast Asia, for example, is facing a crisis of electoral integrity as AI-driven deepfakes (Opens in new window) influence elections. As previously discussed in The Interpreter (Opens in new window), Indonesia, Malaysia and the Philippines each banned Grok over its use in generating deepfakes. ASEAN is now converting that momentum into something more binding in the form of its Digital Economy Framework Agreement (Opens in new window), a treaty-level agreement (Opens in new window) that will embed AI governance provisions.

Regional bodies are far from immune to the power politics that besiege global organisations. A dominant economy may hold outsized influence over which standards are set. But the anticipated dysfunction at the regional level can be contained more than at a global level, while even flawed regional governance models move faster than finding multilateral consensus. As AI tools grow more sophisticated by the day, some form of binding governance is better than none at all.

That speed of action still comes with a fragmentation risk. Different standards across different blocs could create compliance hurdles for AI developers operating across borders. For this reason, cross-regional harmonisation is necessary. This could take the form of shared technical benchmarks, building on existing principles like the OECD AI principles, (Opens in new window) or interoperable standards, which already have strong precedent in the defence and security domain (Opens in new window). With AI becoming a national security issue (Opens in new window), this precedent is worth borrowing from.

Regional coalitions will not dissolve the binary race to govern AI between the US and China, but they offer a more realistic third path, allowing countries to shape AI governance standards without having to fully align with either superpower, and giving them agency on an issue that will define the next few years.

Amodei himself compares (Opens in new window) AI’s geopolitical impact to nuclear weapons. However, there is a crucial difference. With nuclear weapons, much of the world’s arms control governance architecture was built in the aftermath of the Second World War. With AI, the window to get governance right is still open, and regional coalitions may be the fastest way to do it.