An analysis of 857 model releases from nine major Chinese developers shows that only 31 included published safety results, and none addressed dangerous capabilities like loss-of-control or self-replication. The regulatory framework prioritizes application-layer content governance over pre-deployment compute thresholds, allowing rapid iteration without independent verification. This structure creates a visibility gap for financial institutions assessing the systemic risk of unvetted agentic deployments in enterprise environments.
152 other tracked developments touching the same claims.