I halted the deployment after real-time risk telemetry during the canary phase showed volatility and liquidity stress metrics breaching pre-set thresholds, indicating the model could amplify market instability under current conditions. Cross-checks revealed a growing gap between live order book dynamics and our simulation assumptions, plus elevated tail-risk sensitivity that increased the chance of procyclical feedback. Governance checks also flagged an unreviewed parameter change in the execution layer and incomplete validation of the automated kill-switch path. To safeguard market functioning and public trust, we will keep the deployment paused pending independent model audit, adversarial stress tests, and a 24-hour shadow run with hard circuit breakers enabled.