The G7's AI Problem Is That Governments Regulate Systems They Do Not Control

Artificial intelligence creates an unusual governance problem for democratic states. Governments possess the legal authority to regulate companies, but many of the systems they are trying to understand are designed, trained and operated by private firms with far greater technical resources and direct knowledge of their capabilities. June's G7 discussions brought that imbalance into the same room.

Private capability, public responsibility

Executives from major AI companies met political leaders as governments considered competitiveness, safety and strategic dependence. The state can set legal obligations, procurement rules and export controls. Companies decide how models are built and often possess the first evidence of new capabilities or failures.

That creates an information asymmetry. Effective oversight requires regulators either to develop enough technical capacity to challenge company claims or to require testing and disclosure that makes external scrutiny possible.

Geopolitics complicates safety

Governments also fear falling behind competitors. A rule that appears prudent in isolation can look costly if policymakers believe it slows domestic firms while foreign developers continue. Europe adds another layer because much of its AI infrastructure depends on US technology.

The result is a governance race with conflicting incentives: regulate enough to control risk, invest enough to remain competitive and avoid becoming dependent on systems another government can restrict.

The durable question is access to evidence

Grand declarations about safe AI will matter less than who can inspect advanced systems, what incidents must be reported, which evaluations are independent and what happens when a model crosses a defined capability threshold. Governance becomes real when public institutions can obtain evidence rather than merely request reassurance.