2026-03-31: Elevated count of Claude Opus 4.6 rate limit exceeded errors
# Elevated count of Claude Opus 4.6 rate limit exceeded errors (Severity 3) **Problem**: A surge of rate limit errors affected users accessing Claude Opus 4.6 and Sonnet 4.6 through AI Gateway and Duo Workflow Service, triggered by a spike in proxy traffic and abusive trial account usage. **Impact**: Users of AI Gateway and Duo Workflow Service, including Duo Agent Platform chat and LLM proxy endpoints, frequently experienced rate limit errors with Claude Opus 4.6 and Sonnet 4.6. Requests failed when the quota was exceeded, though switching to a different model provided a temporary workaround. After blocking the main abusive IP, error rates improved, but some elevated traffic from other IPs continues to be observed. **Causes**: A single external IP using automated trial accounts sent over 68% of requests to the Duo Agent Platform proxy endpoint, causing unusually high proxy traffic and exhausting the shared Anthropic quota for Claude Sonnet and Opus models. Monitoring did not aggregate usage across AI Gateway and Duo Workflow Service, and missing configuration for Sonnet 4.6 and Opus 4.6 in the vault contributed to untracked usage. **Response strategy**: We published a status page update to inform customers and recommended switching models as a workaround. We deployed a Cloudflare rule to block the abusive IP, which has blocked over 15,000 events and resulted in a significant drop in suspicious traffic. Work is ongoing to implement a permanent fix for trial account abuse, aggregate usage monitoring, and add missing model configuration limits. A quota increase request with Anthropic is still under review. _This ticket was created to track_ [_INC-8876_](https://app.incident.io/gitlab/incidents/8876)_, by_ [_incident.io_](https://app.incident.io) 🔥
issue