The core constraint for this outcome is an unusually strict regulatory conjunction: an instrument must grant a government body the authority to conduct or commission its own capability evaluations of already-deployed models, and it must dictate a strict recurring schedule. As of August 2026, no enacted law or active proposal satisfies both limbs. Current legislative frameworks consistently diverge: they either grant discretionary state power (like the EU AI Act) or mandate developer-retained audits (like US state laws) 55 sources. Mandating a regulator to continuously test deployed models is a massive state capacity burden. Breaking this paradigm will require a fundamental shift, likely catalyzed by a major incident where a deployed model significantly alters its behavior post-release. Given this structural barrier and current deregulatory headwinds 3 sources, early resolution is highly unlikely. The first realistic window opens by March 2029 tracking EU AI Act structural review cycles, but structural barriers to state-run evaluations push the 25th percentile to April 2031 and the median to June 2034. The right tail extends to September 2039 for the 75th percentile, and the 90th percentile reaches "never", reflecting the substantial probability (20%) that governments permanently prefer the cheaper paradigm of ad hoc state power combined with mandatory private audits.