The coordination gap in frontier AI safety policies

Prevention alone is insufficient: Current frontier AI safety policies focus heavily on stopping problems before they occur (capability evaluations, deployment gates, usage restrictions), but lack adequate plans for coordinated responses when prevention fails—a gap the authors call "structural" rather than accidental.

Economic incentives work against coordination: Organizations investing in shared safety infrastructure bear concentrated costs while benefits spread across the entire ecosystem, creating systematic underinvestment in coordination mechanisms. This is a classic collective action problem that requires deliberate policy solutions.

Proven models exist in other high-risk fields: Nuclear safety, pandemic preparedness, and critical infrastructure protection have successfully implemented coordination mechanisms like precommitment agreements, shared protocols, and standing coordination venues that could be adapted for AI governance.

Transparency about decision-making is essential: Effective coordination requires organizations to share not just what triggers responses, but the actual decision-making processes that convert warning signals into actions—what the authors call "note-exchange" of response logic.

Speed of institutional learning matters: Without proper coordination architecture, institutions cannot learn from AI-related failures quickly enough to keep pace with the technology's development, potentially leaving dangerous gaps between when problems occur and when the system can respond.