Claude API incident history

How this is counted

• Source. Anthropic's public incident reporting for Claude. We do not probe, ping or measure the services ourselves, so nothing here is an availability measurement.
• What we count. Minutes of the month that fell inside a recorded incident window for that component — from the reported start to the reported resolution, clipped to the month.
• What we do not count. Severity is ignored: an incident logged as degraded performance counts the same as a full outage. Unreported disruption, regional differences and per-user impact are invisible to us. An incident window is not the same thing as downtime.
• Overlaps. Figures are per component only, never aggregated into one service-level number. Within a component, overlapping incidents count the shared minutes once; the same minute can appear under several components when an incident affected several.