The Classic (v1) conversation and embedding issue is resolved. Recovery was verified at 05:03 UTC on 3 November across production regions using workload checks and successful customer conversations. No further model-permission failures have been observed since the US worker refresh completed at 04:47 UTC.

The permanent permissions fix is deployed. Delayed post-session jobs have been replayed, and the affected queues are empty. The outage history and incident write-up remain available. Follow-up review of the permissions dependency and intermittent failures after the initial mitigation is continuing.

Please contact support if you continue to experience errors.

The issue has been fully resolved, and all systems are operating normally. Thank you for bearing with us.