How Long Was Claude Down, and Why?

For about three hours on Tuesday morning, a lot of daily work just stopped. Claude, Claude Code, Claude Cowork, and Anthropic's API and developer console all went down at once, and thousands of people realized how much of their workday runs through one AI company's servers.

What Actually Broke, and When?

A cascading set of services, not just one. Anthropic's status page first flagged elevated error rates at 14:21 UTC. Within minutes, the company confirmed the problem had spread to the Claude API itself, not just the consumer-facing app. The main status page listed claude.ai, Claude Code, and Claude Cowork as being in partial outage, while the Claude Console and API were marked as degraded performance, a slightly less severe classification but still disruptive for anyone building on top of Claude programmatically.

How Long Did It Actually Take to Fix?

Faster than many recent outages, but not instant. A partial fix rolled out around 14:36 UTC, roughly 15 minutes after the incident began, and Anthropic reported that most existing conversations resumed working shortly after. The harder problem was sign-in: users attempting single sign-on or Sign in with Apple couldn't authenticate, which blocked starting new chats, voice conversations, Claude Code or Cowork sessions, making purchases, or uploading files, even after the core service was partially restored. Anthropic explicitly told users not to sign out while this was ongoing, since doing so risked losing access entirely until the fix was completed. Full recovery was confirmed at roughly 1:12 PM ET, about three hours after the incident began.

Is Claude Actually Going Down More Often Than It Used To?

The headline numbers say yes, but the trend line says otherwise. Anthropic's status history shows 13 separate incidents logged in September 2026 alone, which sounds alarming in isolation. But the company's own monthly uptime figures actually improved across the same period: 99.01% in July, 99.59% in August, and 99.70% in September. In plain terms, outages are getting reported and noticed more, partly because Claude now underpins far more daily business and coding work than it did earlier this year, but overall reliability has been trending upward, not down.

What This Means for Businesses Relying on Claude Daily

An outage like this one is a useful stress test for how dependent your workflows have become on a single AI vendor, a lesson worth remembering from the last time multiple AI services went down at once earlier this year.

• If Claude Code or Cowork failing for three hours would stall your team's work, build a documented fallback process now, not during the next outage

• Avoid signing out of active sessions during a reported incident; following Anthropic's own guidance can prevent losing access entirely

• Track Anthropic's status page directly rather than relying on social media chatter; it consistently posted updates faster than most third-party outage trackers during this incident

Our generative AI services team can help you design AI-dependent workflows with a real fallback, so a three-hour outage doesn't become a three-hour work stoppage.