If you tried to work with Claude yesterday afternoon (US time), you probably ran into trouble. Starting around 11 AM Pacific, reports started piling up: Claude.ai unreachable, API errors, Claude Code crashing mid-session. On Downdetector, reports shot past 3,000 within minutes.
What happened
The outage started around 10:59 AM PT and escalated fast. Claude.ai showed authentication errors, the API returned elevated error rates, and Claude Code broke running workflows mid-session. Anthropic’s own status page confirmed: ‘Claude.ai unavailable and elevated errors on the API’ and ‘Elevated errors on Claude Haiku 4.5’.
In total, more than 12,000 users reported issues — an unusually high number.
The fix
Around 2:20 PM ET, Anthropic announced a fix. Reports on Downdetector dropped steeply afterward — from 3,000 to around 350 in short order. The entire incident lasted roughly two hours.
What it means
Outages happen — at every cloud service. What makes this one notable: it hit peak US working hours and affected all three access points simultaneously (Chat, API, Code). For companies that have built their workflows around Claude, it was an uncomfortable reminder that redundancy plans aren’t optional.
Anthropic resolved the outage relatively quickly — two hours for a major outage isn’t bad. But at a $380 billion valuation and a $30 billion revenue run rate, expectations for uptime grow too. That’s the price of success.
Sources: