Scaling AI crisis diplomacy

Chatham House | Isabella Wilkinson, Alex Krasodomski, Alex Kelati, Ann Meceda, Edward Tsoi, Michael Bąk, Wing-Yee Lau, Roland Bouffanais, Roxana Radu, Shyam Krishna, Anine Andresen, Aaron Halpern, Inbar Bobrovsky, Shahar Edgerton Avin, Rowan Wilkinson

A tabletop exercise on regional AI crisis diplomacy, conducted on the sidelines of the UN Global Dialogue on AI Governance, has highlighted critical gaps in international preparedness for AI-driven emergencies. As advanced agentic and cyber capabilities diffuse globally, the lack of established cross-border backchannels and incident-sharing networks threatens to escalate localized technical failures into broader geopolitical conflicts.

This vulnerability is compounded by geopolitical fragmentation, as the United States and China promote competing, mutually exclusive governance frameworks. Middle powers remain highly dependent on non-sovereign frontier models and lack the infrastructure to orchestrate unilateral technical remedies. Bipolar competition leaves the global governance toolbox empty. To address these deficiencies, this policy report advocates scaling fictional-but-feasible scenario exercises to stress-test readiness, map operational gaps, and build regional capacity before integrating these functions into multilateral frameworks like the OECD Hiroshima AI Process and the ASEAN AI Safety Network.

Comment

The emergence of competing governance architectures, specifically the American Pax Silica and China's World AI Cooperation Organisation, forces middle powers into a precarious position of technological dependency. Without sovereign frontier AI capabilities, these states cannot easily verify the safety of proprietary models or implement unilateral technical interventions during a cross-border crisis. Consequently, regional groupings are increasingly turning to localised frameworks like the ASEAN AI Safety Network to build collective resilience.

This fragmentation of standards will likely degrade the efficacy of global incident-sharing mechanisms during a fast-moving agentic AI failure. As a result, the International Network for Advanced AI Measurement, Evaluation and Science will struggle to coordinate rapid-response protocols across politically misaligned jurisdictions. Ultimately, this institutional friction ensures that emergency backchannels remain restricted to bilateral hotlines between Washington and Beijing rather than a unified multilateral framework.

Strategic Question for Discussion

If middle powers increasingly rely on the ASEAN AI Safety Network for incident response, how will this decentralisation affect the ability of the International Network for Advanced AI Measurement, Evaluation and Science to establish a unified global standard for frontier model safety?

The trajectory indicates that regional networks will create localised safety baselines that reflect immediate developmental priorities rather than global consensus. This bottom-up fragmentation suggests that the International Network for Advanced AI Measurement, Evaluation and Science will function primarily as a technical information-clearinghouse rather than a cohesive regulatory authority. Consequently, global frontier model safety will likely remain governed by de facto standards set by the dominant developers in Washington and Beijing.

Share your assessment in the comments below.

To read in details click here