Effective AI BCP requires validation through exercises and controlled failover tests to prove recovery objectives can be met in practice. Merely documenting backups (Option D), hardening access (Option B), or improving monitoring (Option A) does not confirm that the AI stack—data pipelines, feature stores, model registries, inference services, and dependent infrastructure—can actually fail over and recover within RTO/RPO. AAISM prescribes periodic BCP/DR testing (including model artifact restoration, configuration reconstitution, dependency failover, and data pipeline continuity) to verify readiness and identify gaps before real incidents.