Grok API Records 47-Minute Zero Availability on October 12

Status.x.ai logged the incident as a complete service interruption affecting both chat completions and API endpoints. No external DDoS or infrastructure outage was recorded. Internal telemetry showed the rate-limiter misfiring at 14:19 UTC, triggering pod restarts that exhausted the cluster's capacity until manual intervention restored the limit at 15:09 UTC.

Post-incident metrics released on the status page indicate p99 latency spiked above 120 seconds during the window, with 100 percent of requests failing. Comparable LLM providers report mean incident durations of 18 minutes in 2024; this event exceeds that benchmark by 2.6x. The failure mode matches patterns seen in prior scaling incidents where shared control-plane limits were not isolated from inference traffic.

The outage underscores that production LLM deployments remain vulnerable to single-point configuration errors even when model weights and hardware are redundant. Security implications include potential for similar rate-limit abuse to be weaponized as an availability attack vector without direct model access.

xAI has scheduled a post-mortem for October 18. Expect updated rate-limiter isolation and per-tenant circuit breakers in the next deployment cycle.