International Security Applications of Flexible Hardware-Enabled Guarantees

Hardware-based AI governance could enable international verification: The paper proposes that flexible hardware-enabled guarantees (flexHEGs) built into AI chips could allow countries to transparently monitor each other's compliance with AI safety agreements, moving beyond traditional trust-based diplomacy to cryptographically-verified enforcement mechanisms.

Two complementary governance approaches are analyzed: The report examines verification-based agreements (enabling transparent monitoring of compliance) and ruleset-based agreements (automatically enforcing international rules through signed updates), suggesting these could work together to address different security concerns from AI proliferation to military applications.

Comprehensive chip-level controls could promote strategic stability: According to game-theoretic analysis presented, deploying flexHEGs across all AI-relevant computing hardware—rather than just high-risk facilities—could create stable international agreements that maintain balance-of-power dynamics while reducing risks from uncontrolled AI development.

Implementation requires addressing practical governance challenges: The paper identifies critical obstacles including defining technical thresholds for which chips require controls, managing existing non-flexHEG hardware already in circulation, and preventing powerful actors from abusing governance mechanisms—suggesting that technical solutions alone are insufficient without robust institutional safeguards.

The approach aims to respect national sovereignty while enabling collective security: Rather than imposing external control, flexHEGs are framed as tools that allow nations to voluntarily commit to international rules while maintaining independent verification capabilities, addressing concerns about both AI risks and geopolitical power imbalances.