After Parity
Abstract
An American lead in artificial intelligence may be a strategic objective. It cannot also be the condition under which an international safety regime remains intelligible. This paper distinguishes three arrangements often conflated in the debate: commitments explicitly conditioned on a developer’s lead, domestic oversight that operates within a jurisdiction, and reciprocal verification across jurisdictions. They fail in different ways and require different repairs. A recent US–China policy discussion supplies the starting problem: a lead described in months, attributed principally to compute, and estimated against an incompletely observed frontier. The design test is simple: would a party still accept the rule if the positions reversed? Parity is a condition the regime must survive, rather than a forecast or prerequisite for agreement. Building on the observability floor and the Verifiable Capability Disclosure proposal, it specifies a common evidentiary object: a covered run linked to its configuration, provenance, external access and independently examinable record. It proposes an incident-disclosure clause, graduated inspection triggers, and custody arrangements that do not depend on either country’s permanent primacy. Verification would neither equalize capabilities nor settle acceptable risk. It would preserve a basis for adjudicating claims when the identity of the leader changes. This preprint applies the observability floor proposed in Li (2026), *Which Failure Do You Fear More?* (https://doi.org/10.5281/zenodo.22821194), to reciprocal verification between the United States and China, with an accession procedure for host states such as Mexico.
