Claude's 36-minute outage hit claude.ai, the API and Code
An auth failure spiraled into degraded performance across five services from 21:58 UTC. Anthropic cleared it in 36 minutes — timeline, scope, and your hedge.
Copy markdown
Auth failure, then degraded everything
It began 21:58 UTC on Aug 16 as an issue authenticating users — sign-ins failing, Claude not loading, requests hanging — before Anthropic escalated it to degraded performance across the platform.
The blast radius: five services, API included
Anthropic's status page named claude.ai, platform.claude.com, the Claude API, Claude Code, and Claude Cowork as all affected — so direct API traffic and agent runs went down too, not just the web app.
Cleared in 36 minutes — resolved 22:34 UTC
The incident moved investigating → identified → monitoring → resolved by 22:34 UTC; no root cause has been published yet. If your jobs are still erroring, retry now — the all-clear is official.
The hedge: a fallback route beats a status page
A single-provider dependency means an hour of dead requests every time this recurs — and Claude's outage log shows it does. Front Claude with a gateway (OpenRouter, LiteLLM) plus a warm second model and you just failover.