Voluntary principles have a jurisdiction problem: a company can simply route around them. When one country tightens AI or data rules, development shifts to wherever the rules are loosest, and users everywhere live with the consequences. That is regulatory arbitrage, and it is exactly what happened with data havens and tax havens. The OECD AI Principles are explicitly non-binding, so compliance is a branding choice. Enforceable international rules set a floor nobody can undercut, the way aviation safety standards work through the ICAO. Without a floor, the most responsible companies face the highest costs, which punishes exactly the behavior we want. A binding minimum does not require a world government. It requires countries to agree that some harms should be illegal everywhere.
Should international technology rules be enforceable rather than voluntary?
One FOR case in this debate
Discussion (4)
I keep going back and forth. Enforcement without a real enforcer is fantasy, but voluntary norms have a decent record of hardening into law once best practices settle. Maybe sequence it: norms now, binding rules later.
Sequencing sounds nice but 'later' never comes without a deadline. Norms harden into law only when someone forces the issue, which is exactly what enforceability does.
Regulatory arbitrage is the whole game. Companies already shop for the friendliest data regime, and AI labs will do the same with safety rules. A voluntary principle that a lab can dodge by moving servers is not a rule at all.
The aviation analogy is apt. Nobody argues each country should have voluntary airplane safety. We agreed on a floor because crashes cross borders, and AI failures will too.