Skip to content
Rush Commerce
AI & Automation3 min read

Claude outages on November 13: scheduled tasks ran late

Two Claude outages on November 13 hit sign-in, the API and Claude Code. Scheduled tasks ran more than two hours late and some runs were skipped. Make your automations catch up.

Two Claude outages on November 13 took about four and a half hours of the work day in total. The errors are the part that everyone saw. The part that matters more to an operator is one line in the status updates: scheduled tasks and routines ran more than two hours late, and some runs did not happen at all.

What actually happened

Anthropic's status page shows two incidents on November 13. All times are UTC.

The 15:33 update on the second incident has the important detail. Errors were back to normal levels. But scheduled tasks and routines could run more than two hours late, and some runs were skipped. Claude Security scans were also delayed, and users could not add a new connector on claude.ai for a time.

Neither incident page states a cause, and neither one promises a postmortem. The status page also shows a third item for the same day: errors on the Microsoft 365 connector, which Anthropic attributes to a Microsoft 365 service incident.

Why the Claude outages matter for your business

A chat window that fails is obvious. A person sees the error and tries again. A scheduled job that does not run is silent. There is no error, because nothing started. You find out when the morning report is not there, or when a customer asks about an order that nobody processed.

This is not a Claude problem. Every hosted scheduler can skip a run. We wrote the same thing after a Klaviyo webhook pause. The fix is in your design:

  1. Record each run in your own system. Write a row with the job name, the start time and the result. The vendor's history is not your record.
  2. Alert on a missing run. A dead-man check sends an alert when a job does not report by its deadline. Healthchecks.io and Cronitor do this.
  3. Make each job safe to run two times. Use an idempotency key so that a late run and a manual run do not send two emails or two invoices.
  4. Make each job catch up. The job must process all items since the last successful run, not only the items from the last hour.
  5. Know which jobs need a second route. For a job that cannot wait two hours, keep a second model provider ready.

Key takeaways

  • Two Claude incidents on November 13: 09:58 to 11:58 UTC and 13:09 to 15:38 UTC
  • Sign-in, claude.ai, the API, Claude Code, connectors and Managed Agents were affected
  • Scheduled tasks and routines ran more than two hours late, and some runs were skipped
  • Anthropic's incident pages state no cause
  • Log each run, alert on a missing run, and make each job catch up

An automation that fails silently is worse than no automation. We build scheduled agent jobs with run logs, missing-run alerts, idempotent steps and a second model route. See how we build automation that recovers, or tell us which job you cannot afford to miss.

Sources: Claude Status: Elevated errors on claude.ai, Claude Code, and platform.claude.com, Claude Status: Elevated errors on claude.ai, the Claude API, and Claude Code.

  • #claude
  • #outage
  • #scheduled-tasks
  • #ai-agents
  • #reliability
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.